Context engineering with Dex Horthy

The Pragmatic Engineer · 2026-07-15 · 93 min · 1148 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

1. 00:00So what is context engineering?
2. 00:01It's kind of like deabstracting the abstractions that have been layered on top of rag memory, agentic history.
3. 00:08At the end of the day, they're all different ways to pass tokens into a model.
4. 00:11What is a smart zone and what is a dumb zone?
5. 00:13The less context window you use, the better outcomes you'll get always.
6. 00:17A new paradm that is spreading up is loop engineering.
7. 00:20What do you think is bad about it?
8. 00:22Problem with loops is like at a certain point, you're going to generate so much code that you can't read it anymore.
9. 00:26We built a lights off software factory in July of 2025 and by November we had shut it down.
10. 00:32Can we talk about what you mean by token harder and token smarter?
11. 00:36I'm in a group chat called hyperengineering and it's all like people trying to max out their cloud subs.
12. 00:41That's my idea of token harder and the goal is what happens when you let AI agent ship code for
13. 00:49[music]
14. 00:49months and no developer reads a single line.
15. 00:52Today's guest tried exactly that.
16. 00:53He built a lights off software factory and four months later he had no choice but to shut it down as things just
17. 00:58[music]
18. 00:58stopped working.
19. 01:00Dexory is the founder of human layer and the person who coined the term context engineering days before Andre Carpathy and Tubiluska made it famous.
20. 01:07He spent the last two years talking to hundreds of AI engineers about what actually works
21. 01:10[music]
22. 01:10when you build with LMS and is testing the most extreme ideas with his own team.
23. 01:14In today's conversation we discuss
24. 01:15[music]
25. 01:16context engineering, what it is and the physics of context windows, including what the dump zone is.
26. 01:21loop
27. 01:21[music]
28. 01:22engineering from the Ralph Wim technique to the slow loops that Dex's team runs every night to wake up to code cleanup PRs
29. 01:27[music]
30. 01:28the rise of software factories from a NATO conference in 1968 through DevOps
31. 01:32[music]
32. 01:32to today's agentic factories specdriven development and why specs always drift from the code itself and many more.
33. 01:39If you want to understand increasingly important concepts like concept engineering and harness engineering or want to know how far you can push the let agents build everything idea from someone who pushed it further than almost anyone then this episode is for you.
34. 01:51This episode is presented by antithesis.
35. 01:53If you work with agents your job is no longer just writing
36. 01:55[music]
37. 01:56code.
38. 01:56It's specifying and testing it and antithesis is the most effective method of verifying agenda code today.
39. 02:03Today's episode is brought to you by Buildkite, the CIOS platform trusted by OpenAI, Entropic, Cursor, Nvidia, Uber, Canva, and more.
40. 02:10Today, we're talking about pushing the right context into models so that they write better code.
41. 02:14Right after that starts working, your agents will write more code, a lot more.
42. 02:18Trusting that code avalanche is where many teams face a challenge today.
43. 02:21Every change that an agent makes still has to be built, tested, and proven safe before it ships.
44. 02:26Worked on my machine is not enough.
45. 02:29So, you obviously need CI.
46. 02:30But when agents are pushing five, 10, or 50 times the commit volume into your pipelines, faster CI runners won't save you.
47. 02:37Shaving 30 seconds off a single build is meaningless when the queue is 100 plus jobs deep.
48. 02:41What you really want is a CI system that gets faster as the volume grows and CI that offers instant parallelization to give you unlimited concurrency and to intelligently route changes at runtime.
49. 02:52This is what Bill Kite does and why global software leaders continue to rely on it.
50. 02:56The same architecture that observed the scale of Shopify and Uber a decade ago now runs about 1.4 billion job minutes a week across cursor meta and snowflake.
51. 03:05While the rest of the CI world are cracking under the weight of rearchitecting your platform, build kite continues to reliably grow.
52. 03:10Agents running on your infrastructure or build kites.
53. 03:12Any cloud, any chip, your secrets, your scale, every artifact and log is captured.
54. 03:16So when something fails, either you or your agents have immediate insight for why.
55. 03:20As you're ensuring the context you'll give to your agents, think about how you'll verify what they hand back.
56. 03:24If your system is buckling under the increased volume, head to buildkite.com/pragmatic.
57. 03:2930-day all access trial, no credit card, and an actual human engineer on standby.
58. 03:33His name's Ola, and he's very helpful.
59. 03:36So, Dex, welcome to the podcast.
60. 03:38Super stoked to be here, dude.
61. 03:39Before we get into some of the context engineering and some of some of the the more spicy stuff as well, how did you get into tech?
62. 03:46How did you fall in love with computers?
63. 03:48Oh, man.
64. 03:49So, I uh I was doing undergrad as a as a physics major.
65. 03:53Um, and I realized that uh I didn't like academia.
66. 03:57And there's like basically like two or three paths out of physics is basically you go get a PhD or you go into finance or you go do programming.
67. 04:08At that time, this was, you know, 2012, 2011 when it was like in the middle of undergrad and deciding what to do.
68. 04:12And I had done an internship when I was in high school.
69. 04:15I was working with NASA researchers to a jet propulsion lab in California.
70. 04:20They had just gotten this really highfidelity like the most uh you know fine grain data set of altitudes like the heights of very at very like top topographical map of the south pole of the moon.
71. 04:33And the south pole of the moon is really interesting because some of the craters there are so deep because of the angle it has.
72. 04:38It got hit by meteor storms like no other part of the moon.
73. 04:41So there's very deep craters that have never seen sunlight.
74. 04:44And so there's frozen liquid water in there from the formation of the moon.
75. 04:48And so scientists were really interested in getting down there and exploring.
76. 04:51And uh so we had this really fine grain map and it's like okay cool.
77. 04:54Let's build software so that I I have point A to point B. I know the limitations of my rover can you know max incline up is this, max incline down is that find a path from point A to point B that doesn't like break those rules of the incline.
78. 05:07So I was you know 17.
79. 05:09I had never cracked a CS textbook.
80. 05:10So I wrote I basically like wrote a really naive bad version of Dystra's algorithm for pathf finding.
81. 05:16Uh so I was in college.
82. 05:18I was like, I don't know if I want to do the academics thing, but I really enjoyed programming back in the day.
83. 05:21And so, uh, so I decided to go I got like half of a CS minor and then started working on a API platform team at a software company in Chicago and
84. 05:31Sprout Social, right?
85. 05:32Yes.
86. 05:32And uh, basically never went back.
87. 05:35Yeah.
88. 05:36And then then where did you go from there?
89. 05:38Where did you pick up like the parts of the trade?
90. 05:40Because very early on, your first job that's not really common.
91. 05:43you were doing platform engineering back in you know more than a decade ago.
92. 05:47From that point, it took me about two or three months to notice that like the most valuable work that was being done in the company was being done by like of course it's obvious like the first couple engineers who know everything and understand where everything was and like you spend a day on a support ticket from a
93. 06:02customer and they solve it in 5 minutes but like you have to solve it so you learn and whatever.
94. 06:06And I realized like the most valuable people in the company were the people that were building the developer platform CI/CD sandbox environments preview stuff.
95. 06:16And so I kind of like that was my first step into the journey and I've basically been obsessed with software factories since that like three or six months into my first job.
96. 06:25We talk about software factories now but you're you're talking about software factories back then.
97. 06:29So like you were you're you were starting to already think that this is how we can produce better software inside this is pre AI world right?
98. 06:36Well and I'm always surprised like there's a huge class of developers that say I don't want to work on CI/CD.
99. 06:40I hate CI/CD.
100. 06:41I'm like really because building the thing that builds the thing and building the thing that builds the thing that builds the thing is like as software engineers we're lazy.
101. 06:48We want to do the most high lever thing that makes our job easier.
102. 06:51So how do we if we can build a thing that helps us build a thing that helps us move faster then that's the best use of my time as a as a lazy engineer.
103. 06:58And then you went to another startup uh as aspiration.
104. 07:03Aspiration.
105. 07:03Yeah.
106. 07:04Aspiration also platform engineering.
107. 07:06Yeah.
108. 07:06I was brought in and then like three 3 months into the job, the VP of engineering who hired me quit or got fired.
109. 07:12I don't know.
110. 07:12There was some drama about it.
111. 07:13I probably shouldn't talk about it.
112. 07:14And then I was there for about a year uh and was kind of like acting CTO for a while like hired a couple people, helped hire the new VP of engineering, but I was out of there.
113. 07:22I don't think I'll ever do consumer again.
114. 07:24I think I'm actually a B2B guy.
115. 07:26Good to know.
116. 07:27And then you went to replicator where you spent like a good like like solid like four years and went from engineer for deployed engineer to product manager.
117. 07:34Yeah, I did core engineering for like two years.
118. 07:36We were building a container orchestrator like before Kubernetes, before Docker Swarm was really a thing.
119. 07:41We built our own orchestrator.
120. 07:43The founders had this vision that like, oh, Docker is going to make it much easier to ship on-prem software.
121. 07:48And when I say onrem, I don't mean literally like a a rack in a colo.
122. 07:52It's more like, hey, look, bring the app to where the data is rather than sending the data up to some cloud vendor.
123. 07:57And Docker makes it much more much easier to package up apps and and and and move them around.
124. 08:02And so they had this thesis that like basically you could build a platform that the experience that you get when you use GitHub enterprise which is like you install it and it has this admin panel but then you just get GitHub running in your data center and your code never has to leave your your data center.
125. 08:15Suddenly you could build a generic SAS where everybody could have that.
126. 08:19So I did two years as an engineer there and then our head of sales.
127. 08:23We parted ways with our head of sales and uh honestly I was having a lot of arguments about the software factory with our CTO and it's kind of like almost like a too many cooks in the kitchen kind of thing.
128. 08:31I'm sure many listeners listen listeners have had this experience of like well yeah I know I have these tickets to build but like CI sucks.
129. 08:38I got to fix CI because it's too slow or it's like there's too many different builds and it's always breaking.
130. 08:42like I'm going to fix that and then I'm going to do the end is just like Dex, I need you to stop fixing the build pipeline and like do the tickets I gave you.
131. 08:49I'm sure you've had this experience perhaps.
132. 08:51Yeah.
133. 08:51And then and was this what led you to either forward deploy engineering?
134. 08:56Yeah.
135. 08:56So I like I really loved our customers.
136. 08:58Our customer our customers are Hashi Corp, Data Stacks, Puppet, all these really cool engineering brands.
137. 09:03TravisCI, CircleCI.
138. 09:04I was like yeah I actually love working with our customers.
139. 09:06Our customers are awesome.
140. 09:08And uh it was a great way to like get in the trenches.
141. 09:10a lot of really good engineers who were solving the hardest problem at the company which is like how do we take this 3 to 5year-old SAS platform and package it all up so that someone who knows nothing about our architecture can run it reliably in their own AWS VPC in their own on-prem data center whatever it was
142. 09:27and so I spent I was our first kind of customerf facing engineer and it was in about three months I we closed I met with like every company customer that was like kind of in the pipeline but wasn't moving saleswise is and we closed like 12 deals in 3 months and the CEO was like, "Holy crap, Dex.
143. 09:44Like the the investors are taking my calls again.
144. 09:47Like I don't I know you want to get back to coding, but like I need you to go hire three people and like build this team out cuz I think you might have been like born for this."
145. 09:54Wow.
146. 09:55Yeah.
147. 09:55So I did that for about four years, built that or to like 25 people and then Zer happened and uh it got a lot smaller and we kind of realized like, hey, we have a product that's like pretty good uh and we've been solving
148. 10:05what lots of early startups do is like, okay, there's some usability issues.
149. 10:08is we'll throw we'll get a bunch of smart people, throw them in the trenches with our customers, great for sales, great for retention, all this stuff.
150. 10:14And it was like, oh, we actually like the margins on that aren't aren't good enough.
151. 10:17And so we basically were like, cool, we actually just need to make the product way more usable, do a more PLG-shaped thing, make it productled growth,
152. 10:24product led growth.
153. 10:25Make it a little more self-service so you don't need an expert to teach you how to use it.
154. 10:28And I was like, cool.
155. 10:29If that's the most important thing, then I want to go be a product manager because I have tons of opinions.
156. 10:32I've now spent four years in the trenches with our customers.
157. 10:35I have a laundry list of roadmap things that I think would make the product way easier to use and adopt and implement and deploy
158. 10:41and and now you went the full ar you went towards a dark side.
159. 10:44Exactly.
160. 10:44Yeah, I did.
161. 10:45I was like this is going to kill my street cred isn't it?
162. 10:47But uh I was really glad you know I think a lot of engineers are afraid that if they go do a customerf facing thing they lose all their credibility and like yes I wasn't coding for 10 hours a day.
163. 10:56I was coding for like three or four hours on a Saturday for fun.
164. 10:59Not uh but I mean we were helping people build YAML we were building CLIs.
165. 11:02We owned a lot of the tooling that customers use, but it was like the last mile delivery side of it, not the core platform.
166. 11:09And like on a more personal note, I had spent the last like most of my 20s feeling like okay, a little bit introverted, a little bit like socially awkward.
167. 11:17What I what a lot of engineers I'm sure experience and uh I had talked to my uncle's a music producer.
168. 11:24So he used to work with like Randy Newman and a bunch of like really famous musicians.
169. 11:27Oh wow.
170. 11:28Yeah.
171. 11:28This guy Mitchell F. And he he I was sitting with dinner with him at some point and when I was I think it was when I was still in undergrad, but he gave me this lecture.
172. 11:34He was basically like if you want to be really good at something, you have to make it the only thing you do.
173. 11:40The guy playing guitar nights and weekends trying to get his band off off the ground will probably never achieve greatness.
174. 11:46The people who become great are the people who basically make it like if I don't play guitar, I don't eat.
175. 11:52And you go and you sit on the street all day and you play for 14 hours a day or whatever it is.
176. 11:57That's the only way to become great.
177. 11:59So, I said, "Okay, instead of trying to like read self-help books about how to be less introverted and less socially awkward, like what if I just made it my freaking job to just talk to people and make friends and like help people and solve their problems and uh I think it worked out.
178. 12:12I recommend it.
179. 12:12I think everyone should spend a year or two at least doing something really like customerf facing."
180. 12:16Did you do this because you felt that it was holding you back be being introverted or or like what what what and I I know you got the motivation from the whole musician motivation.
181. 12:26I I get it on one part, but what was it that you said like is a customerf facing thing that I'm I'm going to be doing it because clearly you were pretty great at like writing code by that point.
182. 12:34You could argue you were doing it night and day.
183. 12:36So where where did you find that like I actually I think like customerf facing or like getting this introvert off of me?
184. 12:42Did you feel that I was holding you back or you just wanted to be good at it?
185. 12:45It was just kind of a thing that was like interfering with my like general life satisfaction.
186. 12:50And it was also like I'm not a very type A person.
187. 12:52I'm very disorganized.
188. 12:53is I don't know if people call it like okay I'm like ADHD now that's why I can run 30 quads in parallel or whatever it is but it was like I was really bad at email and calendars and spreadsheets I just like didn't care about these didn't understand them and so like another side effect of this was
189. 13:05like it just forced me to be organized and keep a lot of things going and so like I don't know there's like weird benefits you get from like stepping outside your comfort zone and learning like industrial disciplines that are separate from what you've been doing and so the opportunity presented itself and I was like oh I like
190. 13:18working I'll try this for a little bit started going really well I'm like cool let's keep let's see let's see how far this thread goes
191. 13:24and then afterwards you're now in your second startup.
192. 13:27You you became a founder and you also got involved in in AI pretty early as I as it was even before it was so obvious that it would change how it would change how we develop software, right?
193. 13:38Well, I would say I was I was later than I could have been because we started the company uh me and a buddy in Chicago started a company in the data engineering space in about 2020 November 20.
194. 13:48We decided in like August of 2015,
195. 13:50this is Metalytics.
196. 13:51Metalytics.
197. 13:51Um, technically still the same company as human layer, we just like pivoted the the the mission.
198. 13:56But, uh, yeah, the the the advice I got from every angel investor that, you know, people who just knew CTOs I'd worked for before and stuff, they were just like, look, hitting a lot of heads, wins.
199. 14:05I don't know if you know like the whole DBT data engineering fiverr that whole arc where it was like this huge party and tons of investor money going into all these different companies
200. 14:13and then within by like 2021 2022 there was kind of the zer thing and just this general realization that the TAM for those sorts of tools is not as big as everyone
201. 14:23market
202. 14:24yes the total addressable market for those sort of tools was was not as quite as big as uh as we all thought it was.
203. 14:30Um so it was it was a hard place to raise money.
204. 14:32It was a hard place to get customers.
205. 14:34Yeah.
206. 14:34And then I I met you at while you were at Human Layer NSF at an event.
207. 14:38We you actually talked and we chatted afterwards.
208. 14:41But by by that, this was about a year ago, you were already you you started to have some really strong opinions on using AI.
209. 14:49And one of them was this now famous 12 factor agents manifesto.
210. 14:54Is are we calling it a manifesto now?
211. 14:56I'm I'm calling it a manifesto.
212. 14:57It's a manifesto.
213. 14:58I'm calling it.
214. 14:59Let's talk about this.
215. 15:00This was 12 engineuring principles to build reliable production ready apps.
216. 15:04uh how did you come up with this and maybe we can also talk about some of them.
217. 15:07Yeah.
218. 15:08So um I'll I'll kind of like go to like around August the co-founder I was working with kind of burned out and left and it was very we were on good terms.
219. 15:16It was very mutual.
220. 15:16Um and I decided to start messing with AI stuff and I was building a AI agents and what was really in fog right then was like the lang chain the crew AI these like agent frameworks.
221. 15:26Um and it seemed like there was a ton of you go you go in the crew AI discord there's 10,000 people.
222. 15:30It's like, okay, this feels like the right shape and this there's clearly this eco.
223. 15:34You go in every single one of those projects, they have a Chroma DB plugin.
224. 15:37They have like a Composeio plugin.
225. 15:39There's like clearly like this is the this is the shared interface that everybody is building for.
226. 15:43I say, okay, what's missing from all of this?
227. 15:45The agents can call tools, but it's really hard to like control which tools they call.
228. 15:52And if it's a chatbot, obviously, you can show approved deny in the UI of your application.
229. 15:56But I kind of was obsessed with what I would call like outer loop agents or proactive agent.
230. 16:01Agents that would run in the background, get triggered by events.
231. 16:03I mean, OpenClaw is basically like the biggest manifestation of this of like you have a heartbeat, it wakes up, it sees if there's any work to do, it tries to do stuff.
232. 16:10And my thought was like, I'm not going to trust that agent to do anything meaningful.
233. 16:16if I can't get like a Slack message or an iMessage or something when it wants to do something and kind of guarantee deterministically that I can approve or deny that or deny it with feedback and say actually no do it like this.
234. 16:27So we played in that space for a while and talked to a lot of founders and founding engineers and builders.
235. 16:33We came into YC in the fall of 2024 with this idea.
236. 16:36We're building out this API platform and it was sort of like pedag duty but like it wasn't who's on call to fix the servers.
237. 16:42It was like who's on call to this like routing mechanism for like who needs to approve this agent and can they like escalate it or delegate it or defer it all this stuff and we built it for this ecosystem
238. 16:52crew AI link chain fi there's so many grip tape there was so many in that in that time and then I talked to tons of AI engineers who were actually building really interesting things and like actually making money
239. 17:04doing six figure contracts shipping AI to the enterprise and all of them had tried that stuff for like a month or two and then they had thrown it out and they were just writing all a API calls by hand and they were building more things that look more like pipelines and workflows
240. 17:17than these sort of like hands-off call tools in a loop kind of thing.
241. 17:21And so I talked to a hundred people and I spent a lot of time a lot a lot of time hanging out with one of my best friends uh Vib from uh Boundary.
242. 17:28So they build a programming they built like this like protobuffs for AI thing and they're I think they're about to launch their like full fat like programming language touring complete thing.
243. 17:36But he had this way of thinking about agents and building with models and building with inference where it was a lot more about understanding what structured output really is under the hood.
244. 17:46And every single step in your AI workflow is just tokens in tokens out.
245. 17:51And your job as an engineer is figure out, okay, what tokens do I need to put in to maximize the chance that the tokens out are going to be good.
246. 17:57and kind of distilled all these ideas into about 12 principles and wrote about it on GitHub, posted just like this like 12-page GitHub repo, threw it on HackerNews, got like five, it was on the front page for like two days and it I think it really resonated with a lot of people.
247. 18:10Yeah.
248. 18:10So, I I'll just quickly read the 12 principles and and then let's talk about like one or two that resonate.
249. 18:16Sub 12 are natural language of tool calls, own your prompts, own your context window.
250. 18:22Tools are just structured outputs.
251. 18:24Unify execution state and business state.
252. 18:26Launch pause resume with simple APIs.
253. 18:28Contact humans with tool calls.
254. 18:30Own your control flow.
255. 18:32Compact errors into context window.
256. 18:34Small focused agents.
257. 18:35Trigger from anywhere.
258. 18:36Meet users where they are.
259. 18:38Make your agent a stateless reducer.
260. 18:40The stateless.
261. 18:41Yeah, the stateless reducer one was a little actually someone hit me up on Twitter and uh corrected me.
262. 18:46It's actually it's actually a transducer because there's technically multiple steps in the workflow, but there we go.
263. 18:52But but but of this one, this this was a year ago, so like which is like forever in uh in in how the tooling is is evolving.
264. 19:00Which ones still stick with you where you're like, "All right, these were good that that still seem to hold off."
265. 19:05Yeah, I think I'm going spent most of March writing it, published this in April.
266. 19:09Uh, and then Swix hit me up from AI.engineer and he said, "Hey, can you come?
267. 19:13You want to come talk about this."
268. 19:15So, I gave this talk 12 factor agents in like June 6th, I think.
269. 19:19And, uh, small room maybe like it was packed, but it was like maybe a hundred people.
270. 19:23That was the year at AI engineer where like the lower physically like on on the on the second basement floor was all the super corporate stuff and you go up a level is a little bit more and then like on the top floor is all the like weird cutting edge like startup stuff that like you probably shouldn't
271. 19:36care about yet kind of thing.
272. 19:38So we were up there on the top of this like weird way of thinking about agents.
273. 19:42Uh and then about a week later or two weeks later uh Toby Licki from Shopify says I really like this idea of like context engineering.
274. 19:48And I'm like I I wrote about this two months ago.
275. 19:51This is great.
276. 19:52Toby gets it and then a week later Andre Carpathi is like well I really like I think what we should think about is not prompt engineering but context engineering.
277. 19:59And I was like, "Yes, that's my."
278. 20:00Anyways, I don't know.
279. 20:01If you ask Gemini, depends what day it is, they will tell you either me or Toby or Andre came up with context engineering.
280. 20:08You can't really own a word.
281. 20:09Like I don't no one remembers who invented the word prompt engineering.
282. 20:12But of all the factors, factor three of own your context window.
283. 20:15And basically the only way you can whether it's agentic or a single step at a pipeline, the only way you can impact the quality of your output from AI is by caring a lot about what the inputs and crafting them.
284. 20:26Let's talk about context engineering, which I am going to credit you that you coined it.
285. 20:30I I did some research and like I think you were earlier by a few days.
286. 20:33So there we go.
287. 20:34You you coined it.
288. 20:35We're adding we're adding to the we're adding to SEO juice.
289. 20:37We'll have it in a transcript.
290. 20:39Dex coined context engineering.
291. 20:40Well, and and like a asterisk on that is basically like I learned about context engineering from talking to these hundred engineers and founders.
292. 20:47I just kind of like what was the same about what they were all doing and I put a name on it.
293. 20:52So like I didn't invent doing it.
294. 20:53I was just like I think we I think there's this thing and like vocabulary and names are really important and having like clean ways to talk about the problem especially when like a lot of the content about AI right now is so much hype and jargon that is like meaningless.
295. 21:06I was like okay I think there's a word here that is useful to builders that explains how they should be thinking about building their software.
296. 21:12So what is context engineering?
297. 21:14It's kind of like deabstracting a lot of the abstractions that have been layered on top.
298. 21:17So you have rag, you have memory, you have agentic history, you have structured output, you have all these things that are like different ideas in the frame of agentic programming.
299. 21:28And at the end of the day, they're all like different ways to pass tokens into a model and ask it to produce usually some structured output.
300. 21:36And understanding that is a lot more powerful than trying to learn memory and trying to pick some agent framework off the shelf and some memory framework off the shelf.
301. 21:46I mean, those are these things are all really good.
302. 21:48If you want to get to like 80%, you want to get a really good demo.
303. 21:51But when you have to go from 80% to 95% or 99%.
304. 21:56You need to go down a level and think about what's everything we're putting into the context window.
305. 22:00What order is it going in depending on which model we're doing?
306. 22:03And all of this stuff matters.
307. 22:04You have all of these levers that you can pull.
308. 22:07And it just felt like the right abstraction for thinking about how do I get AI to do the thing I want as accurately as possible.
309. 22:14Why is context engineering started to become more more talked about?
310. 22:18It it was about a year ago.
311. 22:19Was it did it have to do with the the context the the context window that we could pass on to LLMs pretty much.
312. 22:25Did it start to expand or did did we just start to realize that we can do a lot more by passing on from you know the easiest one is of course system prompts
313. 22:34but of course whenever you build an LLM behind the scenes you will pass additional context as well not just to prompt the user you will add a bunch of stuff that's I guess a dirty secret of any any
314. 22:44LM but why do you think the focus is moving on to like all right context is important
315. 22:48I think it always was important I think what had to happen is a ton of smart people again like all these builders I talked to a ton of smart people had to like focus really hard on producing like I want to make software that I can sell.
316. 23:00I want to make something that is accurate enough that I'm proud of and I can sell to an enterprise and they're going to be happy with it.
317. 23:06And there's just like the the the the easiest way to get to really high quality AI applications is by thinking at that token level.
318. 23:15Thinking about a string of different LLM calls like rather than just tools in the loop and it's kind of open-ended and very flexible but not that reliable.
319. 23:24thinking of agents as as workflows, as pipelines, as some mix between maybe a couple tools in a loop versus just, hey, I have my tools and I have my model and I have my system prompt and these are the only levers I have.
320. 23:36And it's actually no, you have way more levers.
321. 23:38It's going to take more work and you're going to have to like understand the LLM with a deeper intuition.
322. 23:42But it was a thing that we always needed and it just took time for people to build with this technology to figure out that like this is the layer of abstraction that allows you to break through the quality ceiling.
323. 23:53And how are cost and context engineering connected?
324. 23:58Yeah.
325. 23:58Um I don't know.
326. 23:59I was I was talking about this with uh someone this morning um about like when you're working with LM, one of the things I I like to say is kind of like
327. 24:06make it run, make it right, make it fast.
328. 24:09see if the world's best LLM at the time I think we did a podcast episode that at the time it was like 03 see if 03 can solve your problem
329. 24:16and then give it to people and see if they want that and then if people want it and you use it a lot then go do a bunch of context engineering
330. 24:24because your engineering time is always the bottleneck like humans trying to figure out and solve problems and build evals and improve and try different dimensions or set up jeep or whatever it is
331. 24:33is always going to be more expensive than just using a smarter model until you have millions of requests a A and then it's like, okay, we're going to do a bunch of context engineering, break this up into three calls, and get it to work on GPT40.
332. 24:44And then we're going to take two of those and make those two work on GPT40 and using old model names.
333. 24:48But the point is like for a certain task in your workflow, can you get GPT OSS 12B, which is like 1/ 1,000th of the cost of Opus, can you get it to solve parts of the problem so that the tokens and the things you're using the smartest frontier models for
334. 25:03are just the things that you really need, that level of intelligence?
335. 25:05But you shouldn't go build all of that and overengineer it until you've proved that you need it that it's valuable that it's like okay this is now I mean we get to Eli Goldrat and like what is the the he had this book the goal right it was about how to model your factory and I'm sure we'll
336. 25:18get to that when we talk about software factories it was like what is the bottleneck in your system and one day it will be latency and cost but it's probably not that when you first start out and context engineering is how you move from the
337. 25:29you you add human effort to the equation to improve the efficiency the speed the price the cost efficiency of your system.
338. 25:36Interesting.
339. 25:37And then one thing that came up more recently and a lot later uh recently is harness engineering.
340. 25:43What is harness engineering?
341. 25:45So I made a post in like October I think about or maybe November of of like hey there's this new thing that I see is like I'm calling it harness engineering.
342. 25:53My definition that I had at the time is not what actually this guy Viv who's at lang chain now does a lot of really good writing on agents and how to think about harness.
343. 26:01He had written something called harness engineering like a couple weeks before me but I hadn't read it at that point.
344. 26:06And my take was basically like okay when you build an agent you use use context engineering.
345. 26:12When you use an agent because we gave this talk in August of 2025 about like how to apply context engineering to how you use coding agents.
346. 26:19And that kind of evolved into this idea of like how do you take a harness like cloud code like codeex how do you engineer against the integration points of that harness.
347. 26:30So commands, MCPs, skills, how you organize your codebase.
348. 26:34How do you kind of optimize the environment that the coding agent runs in to like get the best results?
349. 26:40The same way with context engineer, how do you optimize the inputs to every single prompt?
350. 26:43Well, harness engineering just is like how do I raise the floor so that every single turn of this thing, the results are as good as possible.
351. 26:51And the term got super blurry and some people think harness engineering means building a harness.
352. 26:55And some people think hard harness engineering means building around a harness.
353. 26:58I actually like what Martin Fowler came up with uh as usual he's very good at naming things and he kind of defined the you have the LLM and then you have the inner harness which is like the thing the the tool definitions
354. 27:09and the integration points that like say like a cloud code or a codeex or a amp actually exposes that's your inner harness and then you have the outer harness which is the stuff that you the human do to customize that for your specific needs your codebase your languages
355. 27:23etc that's the best definition I think we have for harness engineering
356. 27:26it's interesting how naming is still so so important, isn't it?
357. 27:31Well, it's like as soon as you name anything, people are most people are I'm actually surprised that context engineering still means the same thing to most people that it did a year ago and that it's even still relevant.
358. 27:40Like that's honestly the the craziest thing to me is like you wrote how many things that were written about AI 15 months ago still matter or still interesting um or are still like
359. 27:51have good advice baked into them.
360. 27:53Stuff changes a I think context engineering has been so long lived because it's it's grounded in the fundamentals of how transformer attention works and until we have post transformer
361. 28:04models or linear attention or whatever it is which who knows when that's going to happen context engineering will be interesting and important to anyone building on AI and can we talk about the physics of of context
362. 28:15uh you you you had a you had a tweet uh this this one the the context reality check this is a graph of uh as you get to 1 million context
363. 28:26just the the quality just drops it.
364. 28:29It goes down.
365. 28:30What do we need to know about like the context?
366. 28:32Again, we we now have models that do have a 1 million context window.
367. 28:35Maybe we'll have even longer ones, but when you start to just put in more stuff into the context, it starts to become less efficient.
368. 28:41Like what what do we know so far in terms of from a practical perspective of like someone who is using the context window to add on a bunch of stuff?
369. 28:48May that be MCP, may that be tools, may that be scales, may that be all of these things.
370. 28:53Yeah.
371. 28:53I mean, so the longer context windows are good.
372. 28:56You can talk to it for longer.
373. 28:57Like they're doing a good job.
374. 28:58But at the end of the day, like especially when you had like Opus, it was like Opus 4.5 and then Opus 4.51 mil or 4.6 and 4.61 mil.
375. 29:08You're not actually getting a like smarter model.
376. 29:12like the intelligence of the model is is what drives its ability to attend to all of the tokens in the context window to figure out on the next turn which parts of this 100k or 200k context window are the most relevant to making the decision of like what is the next tool we call and doing that over
377. 29:28and over again in a loop.
378. 29:30So I don't know there was some study that came out in 2025 which found that and again these are old models so like inflate your numbers but it was like frontier LLMs
379. 29:37can follow about 150 to 250 instructions before it starts to drop off.
380. 29:44Their ability to follow all the instructions just like drops off pretty quickly.
381. 29:48And I think Lori Vos I haven't actually looked at the data but they did a study with like the next generation models a year later and it looks like it's like much better the number of instructions you can get in.
382. 29:57In any case, you have like I split context engineering into like two categories.
383. 30:02You have like the the most people think about like the information budget of like okay I can do rag and I can pull out chunks of this document rather than putting the entire book into my context window.
384. 30:14I can just go grab the pages that matter.
385. 30:16But it's also your instruction budget is like if you give the model too many instructions and especially too many conflicting instructions and that's in your initial prompt and also like if you have a conversation you start going down a path and then you change your mind and you start going down a different you actually I don't want
386. 30:29to do any of that I want to do this.
387. 30:30It's like a it's a lot of computation the model has to do to notice that it has to ignore that whole thing.
388. 30:36And when both of those things are kind of far back enough in the context window that they're only half getting attended to, your likelihood that it's like actually going to like remember the exact instructions you gave it 100,000 tokens ago is like it goes down quite significantly.
389. 30:50This is all very interesting because as engineers there we are expected when you know when we're AI engineers which now a lot of software engineer meaning you just like use LMS to to build software like underneath there's an LLM layer somewhere you're an AI engineer congratulations
390. 31:05but it sounds like the expectation is to be you know to be a good to be a good software engineer preI you need to understand you know how to write good code and it helps when you understand a little bit of the underlying
391. 31:16we didn't need to do that that much over time but it it it never hurts but sounds Like right now we're in this phase that to be an engineer who can write an efficient
392. 31:26AI system that use LMS.
393. 31:29You need to understand the dynamics of the context you need to understand why stuffing your context one way or the other can be compute can introduce latency and all of these.
394. 31:40It sounds like it's kind of more of an intuition and of course there's some understanding but from talking to you you're like well it it it does this computation like I know you know cuz you tried it out right?
395. 31:50Yeah like I'm not I'm not a PhD in machine learning like I couldn't actually go like draw a mathematical proof of how this works but we know attention is quadratic
396. 31:57and the more stuff you put in the more it has to spread this attention out over everything.
397. 32:03This just feels like an absolute new area and like a little bit very different to like what we're used to like software engineering which is like pretty kind of like black and white, right?
398. 32:10That compiles or doesn't compile.
399. 32:12That's true.
400. 32:13I mean there's a different kind of intuition.
401. 32:14I was talking about this earlier as well is like there's a different kind of intuition that you that you develop over years as a software engineer and uh there's many categories of it but the one I'll I'll call attention to that is like a thing that you cannot teach you cannot do you cannot learn in a textbook.
402. 32:28The only way to learn it is like I know bad patterns in software because I have debugged them at three in the morning.
403. 32:34This is my buddy Jake from Netflix said this in his talk at AI engineer code.
404. 32:38It's just like there's no better way to learn what is good and what is bad and what works and what doesn't than suffering through the thing that doesn't work.
405. 32:45Well, speaking of suffering through the things that that doesn't work, uh a new paradigm uh that is spreading up is loops.
406. 32:53Loop engineering.
407. 32:54The idea that instead of writing prompts, just write loops.
408. 32:57Set up your loops.
409. 32:57And this all started with the Ralph Wiggum technique where it it will just well it I I guess that's an early version of loops that were just loops around
410. 33:05and now we're we're hearing with some of the big biggest labs talking about that they're actually just doing looping.
411. 33:11What is your take on have have you done some looping yourself?
412. 33:15Have you set up some loops?
413. 33:17And what do you think is good about it and what do you think is bad about it?
414. 33:22Yeah.
415. 33:22So I think of loops as I mean this could I could ramble on this for 10 minutes.
416. 33:26This is an entire talk, but I'll I'll try to I'll try to lay out some highle stuff and then we can dig in wherever you think is most interesting.
417. 33:32We had Ralph Wickham.
418. 33:33It was actually a year and four days ago was the first time I saw the Ralph Wickham demo and like Jeff Hunley was just like visiting SF and he just like came through and like dropped everybody's jaws with his like, "Yeah, I just ran Sonnet around the clock and spent six grand in six weeks and like I
419. 33:49built an entire Gen Z programming lang."
420. 33:51Look at it compiles and it has a stage two compiler where the compiler for the language is written in the language itself and all the insane.
421. 33:58And the core lesson from all of that I think was the idea of back pressure which is basically and I think a lot of people were doing this for a very have been doing this for a long time which is how do I let the model check its own work?
422. 34:13How do I automate the process of getting feedback into the model?
423. 34:16And there's lots and lots of different flavors of this.
424. 34:19You can have deterministic llinters.
425. 34:21You can have unit tests.
426. 34:22Like part of what made the programming language easy to build with Ralph is a programming language can be infinitely verified.
427. 34:28You write you write the code in the language, you compile it.
428. 34:30If the compiler fails, you go fix the compiler.
429. 34:33You run the program.
430. 34:34If the program fails, you go fix the compiler.
431. 34:36Like it's like it's very very verifiable.
432. 34:39And I think the lesson in loops engineering is like if you can make a problem very verifiable, you can kind of like treat it like a black box
433. 34:50and then have it loop because it will keep improving itself because of the verification loop is already there.
434. 34:55Exactly.
435. 34:56And so like you can do this with CI/CD is like I I do this every time I'm doing a release.
436. 34:59I'm like I'm tired.
437. 35:00The CI/CD is slow.
438. 35:01Cool.
439. 35:01Go research the codebase, make a change, make a pull request, run the test, see if it's faster, try again.
440. 35:07Run the p run the test.
441. 35:08push push to the branch check again see if it's faster and so it's like if it can verify its own work in a loop instead of design instead of saying let's try this approach or let's try that approach or suggest and being really back and forth you just say like my goal is to make CI faster
442. 35:22and you tell the model here's the steps here's the five here's the five steps you're going to write some code you're going to commit it you're going to push it you're going to launch a sub agent to watch the job until it's finished it's going to tell you what happened then you're going to decide what to do
443. 35:34next and so that's that's like the very simplest example I have of like designing loops
444. 35:38and you just set the goal which is cloud code and and I think codecs have both chip/go goal which is you just set the goal and it iterates until it reaches it or or as long as it makes progress towards it.
445. 35:50Exactly.
446. 35:51And so it's like if it's verifiable if you can measure this is auto research too.
447. 35:54Auto research is like hey go make this model twice as fast and like it's just a prompt that tells the model to like go to it over and over again and try things until it actually has good results.
448. 36:03So that's what I think of loops engineering.
449. 36:05I don't know.
450. 36:05We we do a very interesting kind of loops engineering where like the the challenge is like I think it's very easy to get very excited about building the thing that builds the thing or building the thing that builds the thing that builds the thing we talked about.
451. 36:18Uh and so people say, "Oh, we need to like redo everything as this big like aentic first factory, maybe even a dark factory."
452. 36:26And they're like redesigning their entire thing to be their infrastructure for the next 5 years.
453. 36:30And I'm sure one thing we know of in engineering uh and especially uh pragmatic engineering is uh how can you make this more incremental?
454. 36:37How can you make it more continuous?
455. 36:39Uh and a lot of people don't have the option to just hey I ran a Ralph loop for 3 days and it fixed every line error in our codebase.
456. 36:46Here's a 60,000line PR.
457. 36:48Who wants to review it and who wants to sign off on merging and deploying it and uh that there's not going to be any bugs?
458. 36:54Nobody.
459. 36:54So I think the the thing I'm most excited is actually like what we call like iterated loops or like slow loops where we basically have a cron job.
460. 37:01We have the loop the the the structure of the loop is really easy.
461. 37:04It's like run this llinter fix one thing commit and push and then we run that every night in our GitHub actions and we wake up every morning to one PR that makes the codebase a little bit better.
462. 37:15I I like the slow loops.
463. 37:16Yeah.
464. 37:17And it has two dimensions.
465. 37:18So you can add now we have a blueprint for it and actually Kyle just shipped a skill so that you can build these yourself.
466. 37:23you can add more like feedback mechanisms.
467. 37:26So, we have React Doctor for the front end.
468. 37:28We have another anti-attern that has no deterministic tooling, but Kyle's just like, "Here's what good looks like.
469. 37:32Here's what bad looks like.
470. 37:33Go fix one thing and bring it back."
471. 37:34It's like prop narrowing basically.
472. 37:36We have a bunch of optional props and most of them don't need to be optional.
473. 37:38It's like here's how to make the prop not optional so that you know that the code just is like cleaner and easier to reason about.
474. 37:44And so, you can add more conditions, more things of like fix one thing.
475. 37:47I want to wake up to a PR.
476. 37:49So, now we wake up to like four PRs because there's four separate things.
477. 37:52And then the other dimension you can do here is as you gain confidence, you can increase the scope.
478. 37:56Instead of fixing one thing, fix four things.
479. 37:58And so these are like other ways to think about loops where it's like something that's not a human triggers it to start.
480. 38:05Whether it's, you know, an alert from Sentry, whether it's a user feedback like support ticket, whether it's PM writes a ticket, whether it's a test is failing, any of or it's a cron, it runs on a schedule, but it's like the trigger should be something that you don't have to like press a button on and there's a
481. 38:22defined workflow and it makes everything a little bit better.
482. 38:25Dex just described letting agents fix things without a human pressing a button.
483. 38:29But what if a bug is too difficult not just for an agent but also for human to reproduce let alone fix?
484. 38:35This is where presenting sponsor anticysis comes in.
485. 38:37I was recently pairing with the antithesis team where we did a walkthrough of how they helped fix a nasty bug inc the open source key value store used by Kubernetes.
486. 38:45This is a bug that actually happened incd.
487. 38:48The team noticed that the linearization validation assertion failed during the regular anticis runs.
488. 38:53This is not good because the linearization guarantees strong consistency.
489. 38:57So this needs to be fixed.
490. 38:58So what the ETCD team did was run a casualty analysis inside anticysis.
491. 39:03This generates this graph which is a bug probability graph.
492. 39:07Here the x-axis is virtual time and the y-axis is probability.
493. 39:12Now we see that something happened just before virtual time 24 that caused a huge jump in the probability that the bug would occur.
494. 39:19Going deeper, we can look at the entire set of timelines.
495. 39:22Vertical lines going down represent events branching off from the same state and the purple dots are where the buck happens.
496. 39:29If we look closely enough, we see that all of the failures come from one parent branch.
497. 39:33Gotcha.
498. 39:34This is such a useful debugging tool.
499. 39:36In the end, the team was able to figure out that process pauses were causing the bug using all these anticis debugging tools.
500. 39:42This non-deterministic bug was diagnosed in a deterministic way.
501. 39:46How cool is that?
502. 39:47Oh, and this is an actual bug that then got fixed incd.
503. 39:50You can see the bug and the fix in ATCD's GitHub repo.
504. 39:53Honestly, the tools that Antys is built for debugging feel pretty darn futuristic, but they are also really powerful.
505. 39:58Head over to antithesis.com/pragmatic to learn more.
506. 40:01I'd also like to talk about our season sponsor, Sentry.
507. 40:04Sentry is a tool I use for application monitoring on all of my projects, including their pragmatic engine back end.
508. 40:09I've used it for 10 years now, starting with when I worked at Uber.
509. 40:13A neat Sentry feature I'm liking is their SE AI agent, which helps investigate production errors.
510. 40:18For example, here's an actual error I had in my application.
511. 40:20I can just ask Seir what might be the root cause and it brings context.
512. 40:24And it can also make a plan to fix it right from the web interface.
513. 40:27And a nice thing is how Seir also works great from Slack as well, not just from the web.
514. 40:31One place I find even more handy to use Sentry is from codeex or clock code using Sentry MCP.
515. 40:36Also, you can set up neat automations like when a resolve sentry issue resurfaces.
516. 40:40You can kick off a cursor agent or GitHub copilot agent to investigate the regression, read the relevant code, and open a PR with a suggested fix.
517. 40:47I'm not a fan for using AI tools just for the sake of it, but I really like the practical integrations where I can fix errors faster and with more context.
518. 40:55Check out Sentry at centry.io/pragmatic and start monitoring and fixing regressions today.
519. 41:00And with this, let's get back to Dex and to Agentic loops that trigger themselves.
520. 41:05Now, you said we can get more ambitious and we can add more things to it, but I'm I'm going to quote you with uh with one of your tweets
521. 41:12which says, "This may surprise you that this is coming from me, but I think we're in for a 1 to three year period where stuff might break at 3:00 a.m. and you're relying on loops to fix it and nobody understands what's under the hood, and you're looking at an ex existential threat to your company."
522. 41:27Yes.
523. 41:28Uh yeah, that one was great.
524. 41:29That one did a lot of numbers.
525. 41:30Uh
526. 41:31[laughter]
527. 41:32it resonated.
528. 41:33Here's the other side of it is like I think that the today with today's models, today's programming languages, today's infrastructure, you might get away with not reading the code.
529. 41:43Problem with loops is like at a certain point you're going to generate so much code that you can't read it anymore.
530. 41:48This is the strong VM dark factory.
531. 41:50This is like Ryan Leopo's like harness engineering.
532. 41:52Just spend as many tokens as possible.
533. 41:54We tried this.
534. 41:55We built a lights off software factory in July of 2025 and by November we had shut it down.
535. 42:01I think it takes about three to six months of you shipping all the time with nobody reading the code before you realize like, wow, this is getting way worse and it's easier to start over than it is to fix it.
536. 42:10Like the models have made the codebase so bad that it is actually going to be easier to just like rethink this from scratch.
537. 42:15And maybe that's okay because we have AI and it's easier to rebuild things from nothing.
538. 42:18And like usually when engineers say like, "Oh, we can't fix this.
539. 42:21We have to rebuild it."
540. 42:22The feedback is like, "No, just refactor in place.
541. 42:24Just constantly keep the codebase getting better."
542. 42:26You mentioned what I said.
543. 42:27You'll notice what I said was not use loops to ship the features that users want.
544. 42:31We use loops to actually improve the codebase quality and we read all the code because we care about how it's architected and we care not just about the system architecture
545. 42:39but what I would call the program design which I think is something people are going to where are the interfaces where are the seams how are we doing dependency injection all of these things that like make your codebase more maintainable over time and keep you from
546. 42:52falling into this trap of like okay well now if I change something over here I broke something over here.
547. 42:57This is the classic problem of software engineering that like software engineering was invented in the 1970s because we realized we needed techniques for avoiding that problem of like this giant ball of spaghetti.
548. 43:09And I don't think the models are smart enough and I don't think we actually have the training and the benchmarking and the eval techniques to get models to write code that is more maintainable over time versus they're all trained on SWE and SWEBench looking things, right?
549. 43:24All of the benchmarks are basically like here's a commit in Django.
550. 43:28Here's an issue that was filed around that time.
551. 43:31see if you can create the fix that the human created.
552. 43:33And it's Django and it's Apache and it's there's a hundred repos in Go and C++ and Typescript and Java and all these different languages, but they're all it's like the problem with training models on maintainability
553. 43:44is like the cost function of bad architecture and bad program design can't be evaluated by running the unit test because it hits you 3 to 6 months later when you're like, "Holy crap, like no one can make it's this software has become so hard to change."
554. 43:56Is this not similar to how senior software engineers why it took years for someone to become a senior?
555. 44:04Because typically and in some environments you became a you can become a senior faster typically fast moving where there's a bunch of issues and you have to keep fixing it.
556. 44:12Sometimes you know some people are working in the same place for 10 years and they're still not that level.
557. 44:16The point was it it just takes time for you to understand the m the small mistake that you make right now that snowballs into like something disastrous later and
558. 44:25you get hit by it and you realize like okay things like you know like testing matters architecture matters tech depth can actually be a killer you know we don't talk about it anymore but we used to talk about how techdub
559. 44:35kills or slows down companies so badly preai that their competitors can overtake them or they're just like stuck with a 2-year refactor not shipping any new features and the competition
560. 44:45you shifts a bunch bunch of other stuff and now they're ahead.
561. 44:48And I I will say like it is possible that GPT7 will fix this, but if you are turning the lights off in your software factory and you're saying like,
562. 44:57"Hey, you know what?
563. 44:58Like we're not going to read the code.
564. 44:59It's fine.
565. 45:00The models are smart enough.
566. 45:01If we give it the right feedback and just throw enough tokens at the problem, it will keep getting better."
567. 45:05This is what led to this tweet.
568. 45:07that might work, but if nobody read the code in three months and you replace all of your all of your like code review with loops of like, hey, if a user complains, we give it to an agent.
569. 45:15If something crashes, we give it to an agent.
570. 45:17If a if a PM writes a ticket, we give it to an agent.
571. 45:20If a CEO writes an obnoxious essay about what we should be building in Slack, we give it to an agent.
572. 45:24Yeah.
573. 45:24[laughter]
574. 45:25And then you stop reading the code because that's going to produce way too much.
575. 45:28Like, no one can read it.
576. 45:29And like the the the PR reviews become the bottleneck.
577. 45:31So, you replace that with aentic testing and agentic uh agentic code review.
578. 45:35Uh, but none of these things have intuition for software architecture because we haven't trained it in yet.
579. 45:40And so you're going to wake up one day and you're going to have an issue with this happened to us and like we got through it and at the time like it was still worth it.
580. 45:46It was like spent 3 weeks onboarding back into the codebase that we had stopped reading 3 months ago because no matter how much sophisticating expert prompting we could not get Opus, I think it was Opus 4.1 at the time.
581. 45:58We could not get Opus 4.1 to actually find the root cause.
582. 46:01We had to go spend several days digging through the code and figuring out like, oh, there's just actually a primary key that's being routed through this whole thing that needs to be changed to a different type of object and it needs it.
583. 46:11This actually happened to you.
584. 46:11This happened to us.
585. 46:12Yeah.
586. 46:13And when it happened, I was like, you know what?
587. 46:14That sucked.
588. 46:15That was terrible.
589. 46:16But we did it.
590. 46:16We solved it.
591. 46:17And uh it's still worth it's still worth not reading the code for most of the time at the cost of every once in a while I'm going to have to spend two weeks fixing an issue by hand.
592. 46:26And I don't believe that anymore because I think the amount of code we're able to write now is actually like 10xed or 100xed and I think the problem's just getting worse.
593. 46:34So let's talk about software factories.
594. 46:36Yeah.
595. 46:36In your mind, cuz I feel it's an overloaded word, but what do you think of a software factory before AI and now post AI?
596. 46:44Do you know the first definition of software factory the first time it was used?
597. 46:47No. It was a NATO conference in 1968.
598. 46:50Oh, Grady Buch would know about this.
599. 46:52Yeah, exactly.
600. 46:52Yeah, great.
601. 46:53You should ask Grady about it.
602. 46:54They talked about the idea of like, okay, you actually need to build a system of steps and like just like a factory floor.
603. 47:00You have like the coding part and the testing part and the validation part and the integration part.
604. 47:05We had no CI/CD.
605. 47:06we barely had version control like but you needed a factory and then it was adopted by like um Toshiba and a bunch of companies and then the the next moment was like DevOps and you have like this idea of like okay we're going to do CI/CD we're going to automate we're chef and anible puppet whatever all these
606. 47:22technologies is like instead of having dudes running around data centers like resizing discs and stuff or clicking around the AWS console yeah exactly it was like cool we build loops the server hits 90% disc space that sends an alert to Nagios Nagios triggers a chef front chefs makes the disc the disc bigger
607. 47:38feedback loops, right?
608. 47:40This has been around for a while.
609. 47:41And in 2018, I want to say this guy Nick Chalane who was uh he was like the CTO or chie ch ch ch ch ch ch ch ch ch ch ch ch ch ch ch ch ch ch ch ch ch ch ch ch ch ch chief software officer of the air force, he wrote this 100page
610. 47:50essay of hey the DoD needs a software factory,
611. 47:54the department of defense.
612. 47:55Yeah, the the department of defense and the air force.
613. 47:58And he called it the dev sec ops factory.
614. 48:00And he said we need all the things that all of the good enterprises are using.
615. 48:04We need Jenkins.
616. 48:05We need like code quality scanning.
617. 48:07We need security scanning.
618. 48:08We need CI/CD.
619. 48:09We need to be able to ship.
620. 48:11We're shipping once every three months or once a year.
621. 48:13We need to be able to ship every day like all these other companies.
622. 48:16And the way we do that is we actually embrace all these automations and technologies so that engineers are are 90% of the issues are caught by automations instead of people actually like manually checking it or manually reading the code or manually integrating modules together.
623. 48:30Wow.
624. 48:31Talk about forward thinking in in the government.
625. 48:34I know.
626. 48:34Oh no, as I was surprised like oh nice like this is I mean and that was part of it is like hey look we're falling behind in like you know I don't know exactly all the reason but I I imagine also about like attracting really good talent is like hey look if we have like the modern software
627. 48:47stack and we're building things fast and we care about efficiency and we care about people's using people's time well we care about them spending time on the hard parts of the job not manually looking for SQL injections like you could automate that.
628. 48:58So this was software factories pre AAI.
629. 49:00Pre AI.
630. 49:01Now I've heard the term a lot more because of AI.
631. 49:04Yeah.
632. 49:04Is it the same?
633. 49:05Is it different?
634. 49:06So this is really hard to say without a drawing, but I'll try to draw it out.
635. 49:10At the core of a software factory, you have like a source of work.
636. 49:13Most you you can imagine a linear a Jira a the st source of truth your object whether it's a spreadsheet or whatever is you have like what stages is the work in.
637. 49:21Yep.
638. 49:22And prei you would take, you know, you would maybe do some architecture review planning.
639. 49:26You would maybe do some sprint planning and then people would take tickets off the queue and they would go build them.
640. 49:30And then you would make a pull request and people would review it and you would run CI checks and then you would send it to prod and then it would make contact with your users and your users would complain about stuff and that would go to your support team and back into your work tracker and
641. 49:42it would crash and you would have issues and that would go into your monitoring stack and that would go into your tracker and that was your loop.
642. 49:48And then people would take stuff off the tracker based on priorities.
643. 49:51product managers, engineering managers, engineers prioritizing work and then we go and do that and the first change is is like this long wind lot lots of phases and this is also why when like a developer shifts a bug but by the time it comes back to you it might be two or 3 months or even longer and
644. 50:06by the time it get fixed it might be a year or two and you know this is why when you're using a piece of software it's like that annoying bug and you talk with customer support but it's just a very
645. 50:15like long latencies at each each part of the the factory if you will.
646. 50:20Yeah.
647. 50:20And the the step where someone pulls a work item off a queue and starts working on it is, you know, couple hours to a couple days before it actually gets integrated into everything else and touches user.
648. 50:30And that's in a in a in a great world, right?
649. 50:32Sometimes you go build it and then you merge it and then it actually gets released 3 months later.
650. 50:35But we're going to assume we're in a fairly modern like we're somewhere like the a Netflix or a meta where engineers are capable of shipping 100 times a day or a thousand times a day, but it still takes 2 three hours to do the work.
651. 50:46And now with an identic factory, what you do is you take out that person building the thing and you replace it with an agent building the thing.
652. 50:52And so you have orchestration to trigger things.
653. 50:54You have a sandbox, you have an LLM, you have an inner harness, you have an outer harness, which is like the dev environment you build for the agent.
654. 51:01And maybe you give it a browser, you give it a video recorder if you use like things like cursor background agents.
655. 51:05They've kind of built this outer harness around the inner harness that is the coding agent.
656. 51:09And then you make PRs with that.
657. 51:11problem there is that like okay now now it takes 10 minutes to do a build instead of two hours or two days and so now the bottleneck is code review so okay let's throw a bunch of AI agents at code review and let's do agentic testing so that like we can basically catch a lot of the easy
658. 51:25stuff and humans are only focused on the most like important critical core parts of the codebase and then the next level up of your agentic factory is you do the top it's like okay then it gets deployed it goes to prod
659. 51:36and a user complains you just hook your support queue right up to the agent someone complains about something agent tries to fix it and instead of looking at a ticket and then saying okay go send you just close that loop and instead every time something goes wrong you just get a PR and then every time something crashes
660. 51:50in Sentry or Data Dog or whatever it goes into the tracker it gets picked up by an agent and you get a PR this is the ramp inspect thing this is the the only difference is like then you have so much code to review and people say well let's try turning the lights off let's just take all
661. 52:04the human testing and review steps out and we'll say okay cool if users complain then it's broken and if users don't complain and it's working and we're not going to read the code.
662. 52:12We're going to use we're going to treat the whole system as a black box.
663. 52:14So, you said you tried this out uh when it was like Opus Formula and you you built the software factory was running beautifully until it just blew up on your faces.
664. 52:22How do you think of this model?
665. 52:23cuz I I can see an ideal world where it works, but clearly we're not in an ideal world.
666. 52:28Like where do you think we are like and could some of this actually work at some point or you know like like what what progress are you seeing right now and and
667. 52:38what is the the today the situation like how much of this do you believe we can automate or should we automate?
668. 52:43Yep.
669. 52:44So if you know me, you follow my stuff, you know I stand for three things.
670. 52:47Number one is like cutting through the hype and the jargon and going trying things and talking to people who are using things and figuring out which parts of this actually work and are valuable.
671. 52:55Number two, we talked about words.
672. 52:57I try to find and protect useful bits of language because I think it helps us all move forward.
673. 53:03And when you take a useful word like agents or you take a useful word like software factory and then you semantically diffuse it, this is another Martin Fowler word.
674. 53:10You make it mean everybody likes the word and it all becomes hype and everyone starts agents means nothing anymore.
675. 53:16agents could be a chatbot, it could be a Slackbot, it could be a coding agent, it could be tools in a loop, whatever it is.
676. 53:22So, I like to protect important useful words and like help help us all like elevate the conversation out of that hype and jargon.
677. 53:29And then I care a lot about going one level down beneath where I'm generally working.
678. 53:35I think there's always this is the same thing with context engineering is like I was rarely actually going and like building LLMs or understanding or training LLMs but knowing how they're trained how transformers works
679. 53:45informs how you build at one layer up and for the software factory my version of that is I spent the last couple weeks going really deep on uh reinforcement
680. 53:54learning with uh verifiable rewards RLVR which is like this very productionized like it's not like RH RHF is still like fairly academic and pure RLVR are is this like it's a machine in these labs of how we train these models and I'm studying like the benchmarks for coding agents and the techniques for training them and how we
681. 54:14like give it a small problem have it solve it delete the test changes it made revert them apply a test patch see if it passed and then even the frontier this year we have like we can get into this later but like frontier code and marathon these new benchmarks that are supposed to be like better at evaluating
682. 54:29models's ability to maintain a codebase over time and write maintainable code um and they are better But I don't think they're sufficient.
683. 54:36But it's basically this idea that like the only thing that made claude code good was reinforcement learning.
684. 54:42And the dimension along which it got good was like we made a model.
685. 54:46We trained the model and the harness together.
686. 54:48And so the model got really good at calling the specific tools in that harness.
687. 54:53Really good at reading files, writing files, searching for files, all this stuff through doing these problems.
688. 54:57And that was what made it feel so much better than all the other CLI coding agents that came before it.
689. 55:02And so people like, "Okay, that was so much better."
690. 55:04And they're just going to keep getting better.
691. 55:05But it's like it got really good in one dimension.
692. 55:08And the dimension that they're not getting better in because it's hard, expensive.
693. 55:13Maybe we need to like get a lot more creative with how we design these these verifiers and benchmarks is in how do I make code that in three months is going to like improve the productivity of humans and agents, mostly agents, but humans and agents in the codebase instead of making it worse over time.
694. 55:29And so you think that part is just missing?
695. 55:31We haven't seen too much improvement.
696. 55:34I haven't seen obviously no one knows what the labs are doing internally cuz it's all very secret.
697. 55:38But I think if we looking at where the bench the benchmarks tend to reflect where the labs are, right?
698. 55:45If there is no benchmark that can convey to me did this model write code that is going to make my codebase better or worse.
699. 55:52The best we have is I I think frontier code from the cognition team is really interesting.
700. 55:56They have like did the test pass and then they have like two layers of model review.
701. 56:01So they have a judge model that checks okay is the patch the model made similar to the patch that is like the golden answer set.
702. 56:08So even if the model didn't write the exact code that the benchmark was expecting did was it functionally equivalent and the next one is like a like code quality review from another judge model and
703. 56:19like that's better but it's not it's not sufficient.
704. 56:23And this is why I also think agentic code review is like yes it will catch things and it will raise your floor but I don't believe like the model writing the code is the same model reading the code and if you ask a model hey is this code good it's going to be like oh yeah it's great
705. 56:34comprehensive it's got unit tests you've tried this I'm sure and you say okay review this PR that my coworker wrote and tell me everything that's wrong with it I was like oh it has this problem and this problem and this is this is sickopantic and they want to tell you what you want to hear and so like
706. 56:47it's really hard for me to trust a model to evaluate the quality of of of code that's written And so I I I have some ideas on like, okay, can you build a benchmark where the model builds 20 features in a row and maintains the codebase the whole time and it doesn't know what features are coming.
707. 57:02You treat it like a real product team where you don't know what you're going to build next week until you get there and you find out what's most important
708. 57:08and then can we try to evaluate like can we build a problem like that that's hard enough that most frontier models fail by issue six or seven.
709. 57:16Is it fair to say that you know like we've had the software factory like before AI it was just like lots of loop it was like the the PM
710. 57:24giving the ticket to the dev the dev building it deploying to production user customers using it customer support getting tickets and then you creating PM triaging and it kind of goes around like in this loop is it fair to say that
711. 57:39the software factory of how a company a team builds and maintains software that is changing because now everyone's replacing some parts of it, you know, maybe the the least advanced teams will just be devs are starting to use cloud code or codecs to write faster.
712. 57:54They're not spending as much time on there.
713. 57:55Some others are also having the deployment the feedback.
714. 57:58Some some actually have the agents already oneshotting bucks.
715. 58:02So like is it fair to say that that the software factory is just is just changing everywhere maybe at different speeds but everyone I think every team who is
716. 58:10building production software they're like they're frantically experimenting trying and everyone's at a different pace.
717. 58:16You'll have the AI native starters where most of this will have agents in them and you'll have the the laggers who are or more more cautious ones.
718. 58:24They have agents in a few places but not in the others.
719. 58:26Well, and I think that's the key is like if you want to do loops engineering, you should build one loop at a time and you should keep them small and contained.
720. 58:33Basically, I think everything except stop reading the code is really good advice.
721. 58:38Take support tickets and turn them into tickets in your system and then maybe turn those into PRs.
722. 58:43Great.
723. 58:43The advice that I have and like what we kind of like are chasing at human layer is like how can I add another checkpoint in that factory?
724. 58:51So instead of having one human re view point where you're reviewing PRs and sometimes they're 100 lines and sometimes they're a thousand lines but it's quite a lot of effort for especially if it's bad especially if it needs rework.
725. 59:02It's quite a lot of effort for a human to be like okay this is wrong go change it in this way and then you loop back to the agent and then you come with another one and like doing a lot of loops on there once once the direction has been committed to it's really hard to steer off
726. 59:13like you're better off just kind of restarting from scratch.
727. 59:16How do you build like controls and mechanisms around that?
728. 59:19And then my take is like if you do a little bit of human agent planning and like discussion before you hand it to the impletor whether it's I mean planning and specs whatever you want to call it again this is spec driven development is another word that has become kind of very like muddled as far as what
729. 59:35it means but basically how can we spend an hour before we start building so that the PR when we read it only takes 20 minutes because the code is perfect instead of not touching it just literally saying every user reported issue becomes a PR through the loop and then we read that PR and it takes six hours
730. 59:53because there's back and forth and we have to make changes and things.
731. 59:55It's all I'm all about like let's find leverage.
732. 59:58And so you basically you have three options in the software factory world.
733. 60:01If you're going to go all in on aentic software factories, you can turn the lights off and just let everything flow and pray that you don't create too much slop and pray that the next generation of models comes fast enough before you create a giant pile of ash.
734. 60:16you can slow way down and read every PR and read every line of code.
735. 60:20Uh, and then you're only going to really get modest benefits from AI because that becomes I I think you should expect maybe 30 to 50% lift in productivity is kind of what I see when we go into teams
736. 60:30or you can find the right leverage points where humans can actually an hour spent over here in planning can save you four hours in in implementation in terms of fixing and going back and and getting the design right.
737. 60:44And that's what I call like seeking leverage.
738. 60:46If you can find the right leverage points for the agents to guide the work, then you can actually move like two to three times faster while maintaining a like 99%
739. 60:55like accuracy to like if the humans were carefully writing this code by hand, how would it come out?
740. 61:01Now jumping a little bit back to ideas.
741. 61:03I will come back to this.
742. 61:05This was earlier maybe it was last year but you had the research plan implement.
743. 61:09Can we talk about the original research plan implement framework and then also what you've learned about this approach?
744. 61:14what what you got wrong about it.
745. 61:16Yeah, sure.
746. 61:16Yeah.
747. 61:17So, um I mean the first time we talked about RPI was in August of 2025.
748. 61:21Um and it was basically like the research was this thing of like, hey, before you go build anything, go read lots and lots of code.
749. 61:28Use a bunch of sub aent sub aents in parallel, understand all the code.
750. 61:31It was this technique that like worked really well for hard problems in complex code bases.
751. 61:35You just ask Claude uh to do a thing that that's it would read three files and make a change.
752. 61:40It would have no context.
753. 61:41So, you start the research.
754. 61:42You don't even tell it what you're working on.
755. 61:44You just tell it, "Hey, can you tell me how this system works and this system and how they connect together and then you get a markdown dock out and this is the context engineering part is like that would take a 100,000 tokens of context, but you would get a 10k token dock out of it that summarized it.
756. 61:58Then you would start a new context window and you would do planning and the planning would be and actually realize like the plans that we were building last summer were actually terrible.
757. 62:06But it would basically be this long.
758. 62:07You would say, "Okay, now here's what we're building.
759. 62:09Here's the research doc.
760. 62:11build a plan to implement it.
761. 62:12And uh in retrospect, now that we see like everyone is obsessed with how do I get agents to work for longer, I think the reason why in like May, June, July, August of 2025
762. 62:22that a lot of people became really interested in planning was it was a very powerful lever to get agents to work for longer.
763. 62:31If you said, "Build me a B2B SAS for uh burrito delivery," you'd get like a homepage and that's it.
764. 62:37But if you said, "Build me a plan," it would build out this big plan.
765. 62:40And then in the next context window, you'd say, "Hey, here's the plan.
766. 62:42Here's all the changes we're going to make.
767. 62:44Go imple it would actually keep going until the plan was done."
768. 62:46So the plan was a really good way to anchor an agent and remind it that like, hey, you're not done until this is all finished.
769. 62:52So that was the original RPI.
770. 62:53And the plan doc, what was bad about it is it didn't give you leverage.
771. 62:56The plan was every single line of code that was going to change like in diff blocks and like all the new stuff to write.
772. 63:02And so like people would review these plans.
773. 63:05We recommended this.
774. 63:05We told people to read the plans.
775. 63:07We read all our plans.
776. 63:08And then eventually I found myself like I just kind of skimmed the plans.
777. 63:11And so you're not really using it as a way to resteer the agent.
778. 63:14It's just kind of there.
779. 63:15And then you go write the code and there's a crap.
780. 63:18Some people would review the plans and the code and it's like okay well the plan was took you 20 minutes to read and then the pull request takes you 20 minutes to read and they're different.
781. 63:26And so you actually doubled the amount of time you're spending reading code instead of like doing less of it.
782. 63:30You've anti- leverage.
783. 63:31And hang on was spec different development not related to this the one that Amazon Kira for example and and GitHub workflows again a year ago did which was it also it first generated a plan and it had the human review it and then it started to and you could edit it as well and then it went off
784. 63:48and implement this part and it it looked beautifully on the surface.
785. 63:51It it should have worked great but it's tossed into the garbage outside of some m some maintenance projects.
786. 63:56I I think it just didn't work.
787. 63:58like all all the feedback I got, people just stopped using it because it just didn't really work that well.
788. 64:02It just rhymes to the RPI framework a little bit, the original one, right?
789. 64:06Well, so our thing too, like the biggest difference between RPI and specri development and some people refer to RPI as specriven dev because for some people SD all it means is
790. 64:17I use a bunch of markdown files while I'm coding and forget what's in them.
791. 64:20I just specri those are my specs and I'm using them to drive development.
792. 64:23There was this OpenAI researcher who talked about specri dev and like hey stop reading the code just write the specs and treat like the coding part as compiling specs into code.
793. 64:34that part never really materialized.
794. 64:36Maybe with GPT7, you know.
795. 64:38Um, but the challen I'm on a GitHub issue in specit uh that has been open for a year and every couple weeks I get there's a new email on the thread of people complaining about this problem of like, okay, I edit my specs and then I edit the code and then the code drifts and the specs how
796. 64:52do I keep the specs up to date as the code is changing and it's basically like you now have two sources of truth and it's it stops being useful.
797. 64:59And so like that's why when RPI the idea of the docs is they were all for a while we kept them around but after two or three months we're like oh these are actually like tactical execution docs.
798. 65:07I do the research I do the plan I do the implementation I throw the docs out and the next time I need research I just do it from scratch because tokens are cheap and my time is expensive
799. 65:17and the amount of time I might waste if I reuse a research that is no longer in sync with the real state of the codebase.
800. 65:24So we just create it live every time.
801. 65:25This is why it's like context engineering still matters.
802. 65:28Creating artifacts that compress the state of the codebase and compress the intent of the builder into small things that can be reused in the future for the scope of a task
803. 65:38is like a very powerful like tactical approach, but it's not a thing like I I have very few opinions on like what sorts of docs that you should leave lying around your codebase that are like evergreen.
804. 65:49I've seen people try to maintain parody between documentation or specs and the code itself and I don't think anyone actually like found it very useful.
805. 65:57Like you can do it and it works but it's like the ratio of the effort it takes to keep them up to date and the and trivially you could do this with AI probably but I've never known anyone who was like yeah this is great and we're glad we have it.
806. 66:08Like you could do it and it might help but I I don't think anyone found it useful enough to like maintain a system to keep the specs and the code in sync versus just using the code as the source of truth always.
807. 66:17Now you mentioned something interesting which is with context engineering you need to sometimes compact and you've previously co talked about intentional compaction that when context is noisy deliberately compress the useful part into a clear like markdown artifact verify it and then start a fresh conversation.
808. 66:34Can we talk about this kind of compaction and why it's important and and it sounds like it's going to be a building block where it already is for context engineering, right?
809. 66:43Yeah.
810. 66:43No, frequent intentional compaction is the building block.
811. 66:46It is it is completely comes from context engineering is context engineering is like how do we get the most out of today's models?
812. 66:53How do we change what we're putting into the model into the context window into the agentic chat?
813. 66:57How do we control that in such a way that we get the best results possible which means doing as much work as possible in the smart zone the you know first 100,000 tokens of the context window.
814. 67:07And uh this intentional frequent intentional compaction is basically like okay the research step we're going to go read a bunch of code and turn it into a doc.
815. 67:14That's our compaction.
816. 67:15We take that forward in the next session.
817. 67:17We're going to read we're going to read the ticket and the intent and turn that into a design document that we call is like okay here's the highle spec of what we want to do.
818. 67:25Here's a high level like current state desired end state and then a bunch of design questions the model has kind of like a very thorough maybe even overengineered like plan mode.
819. 67:34And then you take the research and the design and you do a new session, new context one.
820. 67:38You're like, cool.
821. 67:39You you've compressed the intent and you've compressed the state of the codebase so that you can then do your planning of like, okay, we know what the end state looks like.
822. 67:46We know where we're going.
823. 67:48Now, let's break down how we're going to get there.
824. 67:50All of these different steps of the process exist because models have shortcomings in each of these phases.
825. 67:56So, the research is pretty hands-off.
826. 67:58I don't read the research docs.
827. 67:59It's just like go read a bunch of code and then like make a doc out of it.
828. 68:02Models are pretty damn good at that.
829. 68:03If you ask it to find a bug and have opinions about the codebase, that's different.
830. 68:07But if you just ask it what is the intent and how do this stuff fit together, uh that's usually pretty straightforward.
831. 68:11But designing the end state of the of the software, the architecture and the program design, models are not great at.
832. 68:18They make a lot of like they make decisions and sometimes they're right and sometimes they're wrong.
833. 68:21So we have want to have a human in the loop there.
834. 68:23And then the steps to get there, I we talked about this before, but models love making what I call like horizontal plans.
835. 68:29If you ask a model like build a plan of steps to go build this app, it's like cool.
836. 68:33We're going to do the database and then we're going to do the services layer, then we're going to do the API and then we're going to do the front end.
837. 68:37It's like, well, that actually kind of sucks because we're going to be on the other side of 2,000 lines of code and let's imagine this is an existing codebase, right?
838. 68:43We're going to make changes to all these different parts of the system.
839. 68:45I can't test it till the end.
840. 68:46And so what I would do is like, okay, how would I have built this if I were building by hand?
841. 68:50Well, okay, I would probably create a mock API endpoint with fake data.
842. 68:53And then I would go kind of get the front end kind of how I want it to look.
843. 68:57And then I would actually go like build a services layer and actually wire the data through.
844. 69:01And then I would make a database migration and make my new table.
845. 69:04And then I would actually add a lot of business logic.
846. 69:06And then I would add a bunch of error handling.
847. 69:09And it's completely orthogonal to how model like models would write the database layer and all the error handling without ever like anyone's ever touched or seen the code or whatever it is.
848. 69:17And so this is another place where we like we like to have humans involved because humans have really good taste and judgment.
849. 69:21Like I would rather read five separate little mini diffs of like things that I can manually verify and explore than read 2,000 lines of code and be like well it's not working.
850. 69:32I don't know where.
851. 69:33You don't know where cuz you wrote the code.
852. 69:34You were supposed to get it right.
853. 69:36We talk about compaction context engineer.
854. 69:37It's like how can you stay in the smart zone of the context window which is again the dumb zone.
855. 69:42I will say disclaimer it's really good training wheels if you don't have intuition about this.
856. 69:46So let's just define these things.
857. 69:49What is a smart zone and what is a dumb zone?
858. 69:51So, it's it's it's a little bit blurriier than like I would like I would like it to be.
859. 69:56I think in November we we talked about the first 40% of the context window, but then we had million smart zone.
860. 70:02Yeah.
861. 70:02Then we had million token context window.
862. 70:03So then I changed it to like the first 100,000 tokens if it's a really like 4.8 I usually will go up to like 200k.
863. 70:09But basically the the thing Jeff Huntley had and Ralph Wickham was like the less context window you use the better outcomes you'll get.
864. 70:15And basically the smart smart zone mean meaning if you have context in that first part it should work a lot better and then like the dumb zone is like once you have stuff there it's kind of forget about it like it'll be confused it's not going to do much like it'll degrade.
865. 70:30Yeah.
866. 70:30And there are times and this is an intuition thing like I will often go up to 3 400k tokens.
867. 70:35Four is rare but I will go up to 250 300k tokens for certain types of work where my intuition tells me that I can keep working without without degrading the performance.
868. 70:44But if you don't have good LLM intuition, like 100K for smaller models, 200K for these like really beefy like Codeex and Opus 4.8 models is usually a good like training wheel guideline
869. 70:56of like if you pass there, your quality of results may be degrading.
870. 71:01The biggest tell I see for this is often the uh model's trying to get the test to pass and your 200k token.
871. 71:07Well, let me try this.
872. 71:08Okay, let me try that.
873. 71:08and it's like trying a bunch of stuff and it's getting more and more extreme and it's like thing oh let me delete your end file and try again like this is where things get really really weird and so it's like if you start to see certain types of if I'm like oh we're at 300k tokens and I
874. 71:22need to like fix the unit test I'm like cool write everything we did to a file or even I'll just do like a a built-in compaction depending on the model and then I'm starting a new session at 30k or 50k tokens and I'm like cool we're going to do a hard thing which is you're going to get
875. 71:35this freaking test to pass and you're not going to be stupid about it by the One thing that you said like about the the the model being dumb is you said that if the model ever tells you you are absolutely right
876. 71:46you should start over and we've all had that when it tells me like oh you know you didn't you're absolutely right and I'm like we just get annoyed but why should we start over what's happening there in your
877. 71:57um observations
878. 71:58yeah that's great yeah and the new the new you're absolutely right I think is uh you're right to push back on that right yes
879. 72:03[laughter]
880. 72:04that's opus right
881. 72:05yeah opus is like you didn't run the test did you right could push back on that.
882. 72:09I totally did it.
883. 72:09But no, for me, you're absolutely right was always what the model would respond.
884. 72:12If you were like, "That's totally wrong.
885. 72:14You did it."
886. 72:15Like you if you if you said something where you were angry or frustrated or just wanted to point out that it's done something wrong, it would respond with, "You're absolutely right."
887. 72:23And most of us have had the experience of it says that and then it continues to do the wrong thing.
888. 72:27So, it's like once it starts doing dumb things because there's there's four things in your context window that matter.
889. 72:33There's like the size of it, how many tokens?
890. 72:36There's like the quality of the information is like is there any incorrect information?
891. 72:39Like if the model had some thinking trace where it decided the wrong thing was true.
892. 72:42Is there missing information?
893. 72:44Does this like have context missing that it should have?
894. 72:46And then there's the trajectory.
895. 72:48And the trajectory is very subtle, but you may have had sessions.
896. 72:52The trajectory meaning you're prompting
897. 72:54the actual history of everything.
898. 72:56I call it trajectory is like the actual history of like what the agent has done in the past.
899. 73:00And so if I say, "Hey, make this change."
900. 73:01and the agent makes the change and then it runs the test and then they're broken and then it fixes the test.
901. 73:05I have very high confidence the next change I asked it to make, it's going to follow that path again because it's like, okay, here's a conversation and the last time the user asked me to do a thing, I made the change, I ran the test, test broken, fixed the test, and then I told the user.
902. 73:17But if I say make a change and it makes a change, it doesn't run the tests, then I'm on a different trajectory.
903. 73:22And if I say, okay, make another change, it's like basically the they're auto reggressive.
904. 73:26So they're they're predicting the ne what's the next message in this conversation.
905. 73:29And so the example we we talked about in uh No Vibes Allowed was of course the like hey the model makes a mistake and then you yelled at it and then it made another mistake and then you yelled at it and then it's like cool what's the next message in this conversation.
906. 73:40Well look if I read the history I should probably make another mistake so the human can yell at me.
907. 73:46So I was like okay that's a great that's a great example of like uh time to start over.
908. 73:50Let's talk about some observations on how software engineuring is changing.
909. 73:53One thing you talked about recently on the evolution of the coding meta is going from token harder to token smarter.
910. 74:01Can we talk about what you mean by token harder and token smarter?
911. 74:05Yeah.
912. 74:05So token harder is I mean I'm in a I'm in a group chat called hyperengineering and it's all like people trying to max out their cloud subs.
913. 74:12Oh wow.
914. 74:12Okay.
915. 74:13It's just like
916. 74:14[laughter]
917. 74:15that sounds like a fun is it fun place?
918. 74:16It's a fun place but it's like all token harder.
919. 74:19It's like look at all the side projects I built.
920. 74:21It's look at everything that uh I I I've gotten my Claude token.
921. 74:25I've got six six cloud code accounts.
922. 74:27I've gotten all of them maxed out every 5 hour period.
923. 74:29I've timed it out so I always use all the tokens and it starts up immediately when the limit resets.
924. 74:35And so it's like I mean getting into Eli Goldrat and the goal is like optimizing for utilization and efficiency of one node in your factory rather than the end to end goal of like how do we ship value and things that people like that are stable and like will last a long time.
925. 74:48But that's my idea of token harder and it's the same thing with the dark factory thing is like hey if you if you if you remove humans from code review you can push more tokens through the system.
926. 74:56So we talk about software factories but what is the dark factory?
927. 74:59Ah so the dark factory is this comes from this idea of like there are factories where uh everything is automated by robotics.
928. 75:06So you can imagine like a car factory where it's all robots building the cars and they don't have lights because there's no humans.
929. 75:13Oh, so that's where it comes from.
930. 75:15The dark factory.
931. 75:16Yeah.
932. 75:16You walk in there's no lights.
933. 75:17There's not even light switches.
934. 75:18So, it will be the fully automated software factory where it it it will be like no human input basically.
935. 75:23No human input.
936. 75:24Raw materials go in, cars come out.
937. 75:27Yep.
938. 75:27And I think in in a micro like you can have many loops that are dark in your in your thing of like, hey, if uh if the code review agent comes back with a problem, you loop that back to the builder agent, it fixes it and comes back and that's dark.
939. 75:38You don't need a human loop for that.
940. 75:40But the full dark factory where you don't read any code, yeah, it's a good way to maximize your token utilization.
941. 75:45And it's like if if your belief is like my job is to extract as much intelligence out of the machine god as I can because that's how I get the most value and the most leverage on my time then token harder.
942. 75:57Um and my take is basically what we talked about before token smarter is like okay how do I move faster?
943. 76:03How do I get as much value out of as AI as I can without having to turn the lights off while still maintaining control and taste and judgment and understanding the system architecture
944. 76:12and having a lot of like applying my hard one opinions through 10 years of software engineering to the design of the program so that I can feel confident that the code's going to get better and more maintainable over time.
945. 76:24It's the same thing of like you look at like the S sur team inside Google.
946. 76:27They brought out this book SR site reliability engineering and the whole take was like hey we're going to go from one data center to five data centers and we need the same sixperson team to be able to manage five data centers and we need the same sixperson team to be able to manage 50 data centers next year
947. 76:42and it's basically how do we apply software to this problem so that instead of scaling linearly of like okay every data center needs five devops people so we need to scale the people with the things how do we continually automate the parts that we don't need so a little bit orthogonal and maybe even like contra contradictory to
948. 76:57what I just said, but this idea of like how do you find leverage and the way the way well I I think what you were saying there is like when Google did that never seek to remove those SRE
949. 77:08from the process at all.
950. 77:10They just said like look can we think ahead and scale yourselves and they actually grew the team.
951. 77:14It wasn't actually six people.
952. 77:16It was more like I think Google specifically said, "Okay, we have five data centers.
953. 77:20Next year we'll have 50.
954. 77:21There's six of you.
955. 77:22We do not want to have 60 people.
956. 77:24We don't want and and then management layer and all that.
957. 77:26It's like how can we do it with like 12 or like or like 10 and then when we'll have 500 and now actually their SRE has grown but but
958. 77:33of course yeah
959. 77:34but but they never you know I I think as engineers like we feel pretty threatened when someone says like all right we just want to have zero engineers like I mean that's not a fun place to work at but
960. 77:44what it sounds like
961. 77:45it's not a possible place to work at if they have zero engineers neither of us can work there right
962. 77:49but do understand the token smarter is like let's keep humans in the loop let's keep adding value and figure out what are the parts which are not as relevant,
963. 77:58boring, where we don't need it.
964. 77:59And so like one developer can probably do more than before, but you are built to like be part of this whole thing and the lights are on in a factory.
965. 78:07Yeah.
966. 78:08And it's like basically I think I think what I'm trying to get to is like the connection here is like S builts a thing where like headcount scales at like a square root function or a logarithmic function whereas their output scales like linearly
967. 78:20and you want the same that the way you do that is with good architecture and good program design.
968. 78:24And so in order to like avoid this problem where you have to throw more people or more tokens at at the problem, if you design good software in such a way that it gets more maintainable and more scalable over time
969. 78:35and like just today it doesn't feel like like basically you need humans in the loop to be able to do that.
970. 78:40Let's talk about uh AI slop.
971. 78:44At one point you wrote, "Yeah, AI can write your code, but it can also write your specs and PRDs."
972. 78:48But the same the same rule is always slop in, slop out.
973. 78:51If you outsource your thinking, you're gonna get garbage.
974. 78:54Yep.
975. 78:55Um, so yeah, that's basically the idea is like the way we think about like getting high quality outputs is like yeah, you could write the code by hand or you could sit with a model and work back and forth and go maybe a little bit faster and you have control and every time it makes a change, you
976. 79:11go read the change and if it's bad, you tell it, nope, we want it like this and you kind of incrementally slowly.
977. 79:15This is like kind of the stage two or stage three version of working with agents where like the agents writing all your code, but you're kind of very much in the loop.
978. 79:23And this will make you go faster, but it won't make you go that much faster.
979. 79:26It won't make you go anywhere near there's like there's like that level and then there's like the maximum speed you can go while still caring about the code.
980. 79:32And then there's like the maximum speed you can go if you turn the lights off.
981. 79:35And so we always think about it as like in terms of leverage is like, okay, let me take everything starts with like a sentence or a voice note ramble like I want to build this thing.
982. 79:42is going to work like this or whatever it is to let's say like on average like two sentences I got to fix this thing or there's a support ticket I got to fix this thing if you can turn that with AI into a one pager and then turn that one page and make sure that's correct and then
983. 79:54turn that one pager into a three-pager and make sure that's correct and then turn that three-pager into a 10-page like detailed outline then you can write a 100 pages worth of code and it's maybe not perfect you shouldn't like sweat over these documents and make sure they're perfect but you're increasing the chance
984. 80:10that like you're decreasing the uncertainty of the outputs.
985. 80:14It's like you can think of like you have like a line of like where it's going and then you have like the probabilities of where like it might go in that range
986. 80:21if you are kind of reviewing along the way as you get more and more detailed into how what you're building and how you want it to be built.
987. 80:27You kind of collapse the uncertainty and the set of end states that you could land in.
988. 80:33That's me doing the physics thing of like you got to superimpose all these probabilities and like I don't know I have this thing that like I think people who really like playing real-time strategy games uh are probably going to be really good with AI because you kind of have to like
989. 80:45I don't know.
990. 80:46Matt Matt PCO was just talking about fog of war and like things that are at the frontier of like there's stuff we don't know about this problem yet.
991. 80:53How can we find that out and how can I make the best decision now knowing what I have seen?
992. 80:58there's a I've seen a couple pieces of information and so there's a 30% chance it's this and there's a 40% chance it's this.
993. 81:04How could I get more information?
994. 81:05So in my head I can like recalculate those probabilities and decide what's the most likely path that's going to lead us to success.
995. 81:11Speaking of the most likely path that leads you to success, let's talk about your company that's you've you've just come out of stealth human layer.
996. 81:20What is human layer and what is the probability that you're setting up for success?
997. 81:24That's a good question.
998. 81:25100% 100% probability uh maybe 110 but uh no uh so human layer is it's an AI IDE it's a collaboration platform and it is building blocks for your software factory and the basic pitch is like engineers solving hard problems and complex code bases basically
999. 81:42there's two categories of builders there's like vibe coders building side projects and then there's people building production software where the stakes are high and if something breaks we're going to get fined millions of dollars or you know we're going to lose millions of dollars of money for the company and there's a whole spectrum in between there.
1000. 81:56But it's like if you're kind of in the left half of that spectrum, you're building software that matters and it has to last and be around for a while,
1001. 82:04then you're helping people like that solve problems two to three times faster without descending into slop is like how do you maintain that near human level of quality and move two to three times faster?
1002. 82:14And what were the ideas that you you built and that you came with?
1003. 82:18one idea that we're really excited about right now.
1004. 82:20I mean, it all comes from this RPI and this like using specs to like I mean I've kind of been hinting at it this whole time, right, of like okay cool like start really high level and zoom in layer by layer and resteer and like find find that leverage that helps you move faster and increase the chance
1005. 82:35that your agent's going to build exactly what you want or something that's really high quality.
1006. 82:40The other thing I think that's really interesting that where I just posted yesterday I said, "Hey chat, should we uh kill the poll request?"
1007. 82:45And that's uh something I can't talk too much about, but basically the idea is like the IDE of the future needs to be rethought from the ground up for agents.
1008. 82:54And it might not even be a like I don't know a lot of editors kind of started with the text field and bolted on an agents tab.
1009. 83:01And then eventually you've seen like cursor 3. I can't even find the text editor.
1010. 83:04I know it exists.
1011. 83:05People have told me you can get to a text view of files, but it's also very agent first.
1012. 83:09And so we started from the ground up of like what is an IDE designed for helping a developer interact with and manage the work of agents.
1013. 83:17And then we zoomed out and said how do we make this collaborative and build in a sync engine and durable streams and all of these like pieces of tech that enable
1014. 83:25me to get human input and feedback on what I'm doing with agents in real time rather than waiting for the pull request time.
1015. 83:32And great engineering teams have been doing this for decades of like, hey, we're gonna have a design review where we're going to talk about how we're going to build the thing as like a two-page Google doc or whatever 10page what, however,
1016. 83:42BRD er architecture requirements document and then you go to sprint planning and you break it down into little tickets and you decide who's going to do what.
1017. 83:50It's like AI can help with all of this.
1018. 83:52You should, if you're just using AI to write the code, you're missing out on a lot of the benefits that AI can bring to your SDLC.
1019. 83:58And a lot of people say like, "Well, we don't need any of those meetings anymore because we have the loop.
1020. 84:01We have the dark factory.
1021. 84:02Things just fly around the loop."
1022. 84:03But it's like, "Okay, but if you want to actually move faster and maintain quality, then like you should have these checkpoints before you go to actually write the code and you should use AI to help with that."
1023. 84:11So, we built this like cloud platform that's kind of has like a Google Doc style component where you can comment and the agent can surface like mockups and mermaid diagrams and HTML and all these things.
1024. 84:20So, basically, how do we make agents like Big Figma style?
1025. 84:24Everything's in the cloud.
1026. 84:25Everything's collaborative.
1027. 84:26I see all my co-workers sessions.
1028. 84:27they see all of mine.
1029. 84:28It's almost like the benefit that Slack had over email was that you didn't have to be in every conversation to know what was happening.
1030. 84:37You could maintain you could see all these channels light up.
1031. 84:39You could check on them.
1032. 84:40Okay, I don't care about any of that.
1033. 84:41But if you saw a conversation that you cared about, you could jump in on that.
1034. 84:44And it's like how do we do that for engineering work versus like we really had these like very strict even when we called it agile it's very waterfall like
1035. 84:53PRD ard tickets everyone goes and builds for a day and then you get the PR back and then one person reviews it.
1036. 85:01How do you create this more just like soup and like what is the data model for that world where you have like agentic traces, you have documents, you have tasks and projects that group these things, you have actual git diffs being streamed everywhere where it's like why would I review all the code at once
1037. 85:17when I can just always every everybody's work lives in a shared environment that anyone can go interact with.
1038. 85:23I mean what it reminds me is like what you know GitHub that did the software teams before GitHub and its competitors you might have a tracker somewhere but most teams were just kind of like in inside the company you didn't know what one one team was I I remember pre- GitHub like you know you had individual teams
1039. 85:41they some of them had like a board with stickers but no one else in the company knew what they were doing they were all working isolation and now when you have GitHub or even the internal version of of GitHub inside a company you can always see when when you go to a team you you see the pull
1040. 85:55request flying you you can join in you have history it's all it is all kind of connected and it it came together and now it's like you know for a very long time I was like you duh you're going to use GitHub or or people will copy it so do I sense that you're trying to build something
1041. 86:08like this this workflow for like when you have the the software factories which are like dark factories and loops at a bunch of places how can we have this this new way of working which which will feel natural but like coming up with it like is is hard work and it's it's counterintuitive.
1042. 86:25How can we do something that accomplishes what GitHub did but like 10x better like more specifically like more continuous and more real time and more collaborative than like these discrete units of work that is like the poll request.
1043. 86:37Well, I now I'm starting to understand why you're saying maybe we should kill the poll request because pull request was invented by GitHub, right?
1044. 86:42like it's it is not part of Git, but they did it as a way for you to do a code review merge before it goes in and be able to modify it or or like just reject it, etc.
1045. 86:53And it's probably a lot better than whatever we had before, which I guess was like emailing your git patch to Lionus and ask him to merge it into the kernel or whatever.
1046. 87:00They still do that.
1047. 87:01It work it works for them.
1048. 87:02That's the point.
1049. 87:02But it only works for them.
1050. 87:03Yeah.
1051. 87:04I don't know anybody else who does that.
1052. 87:05I mean, I'm sure even before Get Up for you, you guys had what, like CVS or
1053. 87:08CVS?
1054. 87:10So, if you had a lot of money for Microsoft,
1055. 87:12they made us use subversion at in undergrad because the guy who invented subversion uh was a you Chicago guy.
1056. 87:19The year after I graduated, they switched everybody to Git.
1057. 87:21And I was like, damn, I learned a useless thing just for somebody's ego.
1058. 87:26specifically for AI startups or startups building on top of AI or building AI products.
1059. 87:30How important do you think location and network is especially you are based in in the the valley we see research that AI startups are more frequently funded from here than normal startups as well.
1060. 87:43Do do you see this advantage and also do you see some disadvantages of being a specific may that be Silicon Valley or elsewhere?
1061. 87:49I don't have really strong opinions on this.
1062. 87:51Actually, like Paul Graham gave a talk in Sweden about why SF is cool.
1063. 87:55Rather than just regurgitate that, I will I will forward people onto that one.
1064. 87:58Um, we can put it in the show notes or whatever, but he talks about all of the dynamics of Silicon Valley and the pay it forward culture and the like people take you way more seriously just because you're based here.
1065. 88:08I lived in Chicago for a long time.
1066. 88:10I have a lot of really good friends from high school, from college, from growing up in LA.
1067. 88:13And never before have I felt like so locked in with like my people more.
1068. 88:19Never have I felt more seen, more connected.
1069. 88:22Like there's just so many people here.
1070. 88:23Again, talking about the founder thing, people who care deeply, who are incredibly competent, who like we have all the same types of problems.
1071. 88:30We love all the same types of things.
1072. 88:31Like I don't do land parties where we play video games, but all my buddies will come over and we'll sit in the office till 11.
1073. 88:36We'll just do co-working and like hack on cool fun projects and stuff.
1074. 88:39And like you can't do that anywhere else.
1075. 88:42There's not enough like uh critical mass for that to just happen organically everywhere you go.
1076. 88:47and and I absolutely love it.
1077. 88:48I wouldn't trade it for anything.
1078. 88:50Yeah, I think a critical mass nails it on on the head.
1079. 88:53When it comes to hiring, what types of folks are you hiring for specifically?
1080. 88:57Cuz I'm interested in how hiring changes and and what what a standout engineer means for you and how you are trying to, you know, confirm that those traits exist.
1081. 89:06In general, we we are looking um for people who are have really strong software fundamentals.
1082. 89:12So, understand distributed systems, understand like the core fundamentals of CS and operating systems and these kind of things.
1083. 89:19I mean, you don't have to be a PhD in freaking kernel design or whatever, but it's a lot easier.
1084. 89:24We can we can teach we can teach somebody, I think, to be a really good AI developer in a few months.
1085. 89:28You can build enough intuition where you are, you know, accelerated off the ground and you can go like keep growing there.
1086. 89:34It's really hard to teach someone a CS undergrad program in in 3 months.
1087. 89:38And what's a problem space that you're excited about in in software engineering or even product engineering or building products that you think in the next few years is going to be one of the interesting things that you're going to be attacking?
1088. 89:49My co-founder could talk more about this, but like there's a lot of interesting things happening in in real time in cloud and sandboxes in sync and kind of like
1089. 89:59using these new building blocks that have gotten really solid in the last couple years.
1090. 90:03We're big fans of the electric SQL team.
1091. 90:05where users have durable streams.
1092. 90:06It's like how can you build systems that kind of are a lot more spread out and distributed and almost like decentralized.
1093. 90:14This is really interesting for coding because you want to be able to run coding agents anywhere.
1094. 90:16You want to be able to run them for a short time, for a long time, on demand, on a schedule, all these things and have them all be part of this kind of like
1095. 90:24brain.
1096. 90:25So I don't know, parts of what we're doing are really boring like all our data is in Postgress and then parts of what we're doing is really interesting.
1097. 90:30Um, but there's a lot of distributed systems problems.
1098. 90:32There's a lot of infrastructure problems.
1099. 90:33Like we are building tools for AI, but there's a lot of problems in building collaboration platforms that are really really hard and there's a lot of new tech that makes it easier and more interesting, but it's still uh by far from an easy problem.
1100. 90:47It sounds like what you're saying is like the infr layers to some extent a new infrar being built and it'll take some time and but it'll be like just new new blocks and it will eventually become the primitives like for cloud we have primitives already but it took freaking decade to get those together or more.
1101. 91:02Yeah.
1102. 91:02You had AWS in what like 2008 2006.
1103. 91:06Yeah.
1104. 91:06Uh and then you got Kubernetes a decade later.
1105. 91:09Yep.
1106. 91:09And as closing, what's a book or or reading that you would recommend?
1107. 91:14Something that you personally enjoyed.
1108. 91:15Nowadays, we talk a lot about refactoring by Martin Fowler classic.
1109. 91:20I think it's because we spent a lot of time uh improving the design of existing code and trying to figure out how to get models to build code that is easy to maintain and like easy to read and easy to understand and easy to to build on.
1110. 91:31I feel like I probably have a better answer than that, but that's that's what's top of mind these days.
1111. 91:35Uh we're reading a lot of like classics of software engineering.
1112. 91:37Refactoring clean code, the pragmatic programmer, like all that stuff is I think is more relevant now than it has ever been.
1113. 91:43Love it.
1114. 91:44Well, Dex, thanks so much.
1115. 91:45This was fun.
1116. 91:46This was a blast, dude.
1117. 91:47Thanks for thanks for having me on.
1118. 91:48This was great.
1119. 91:49Uh I had a lot of fun.
1120. 91:50I don't know about you, but I really enjoyed this conversation.
1121. 91:53Dex is such a big believer in gender coding.
1122. 91:55Yet, he's the one warning us that if you stop reading the code, you have about 3 to 6 months before your codebase becomes easier to rewrite than to
1123. 92:03[music]
1124. 92:03fix.
1125. 92:03And this comes from first hasn't experience.
1126. 92:05His team built a light software factory, ran it, and then had to shut it down.
1127. 92:10I also like the idea of the slow
1128. 92:12[music]
1129. 92:12loop.
1130. 92:13Loop engineering feels like a somewhat meaningless term to me.
1131. 92:16What Dex's team does is actually pretty boring.
1132. 92:19A cron job runs every night, fixes one issue or one anti-attern, and opens one small pull request.
1133. 92:25The team wakes up to a codebase that's a little bit better every morning,
1134. 92:28[music]
1135. 92:28and dev still needs to review and prove it.
1136. 92:30This is a practice that honestly any engineering team could just adopt today.
1137. 92:34Finally, I really enjoy the history lesson.
1138. 92:36The term software factory comes from a NATO conference in
1139. 92:39[music]
1140. 92:391968.
1141. 92:40The idea of software used to build software with analogies to a factory is more than 60 years old and every generation of our industry has tried to automate more of the loop of building software.
1142. 92:51AI agents are just yet one more attempt, although probably the most successful one.
1143. 92:55Do check out show notes below for the related the pragmatic engineer deep dives that go even deeper into AI engineering and other related topics.
1144. 93:01If you enjoy this podcast, please do subscribe on your favorite podcast platform and
1145. 93:05[music]
1146. 93:05on YouTube.
1147. 93:06A special thank you if you also leave a rating on the show.
1148. 93:08Thanks and see you in the next