WEBVTT

NOTE Sentence-level transcript of https://www.youtube.com/watch?v=8G_1-3IO4ZQ

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/8G_1-3IO4ZQ.html. A cue ends where the next begins, or 2 s after its last word.

s1
00:00:01.309 --> 00:00:03.309
[music]

s2
00:00:12.640 --> 00:00:13.880
Hi everyone.

s3
00:00:13.880 --> 00:00:14.760
My name is Prakalpa.

s4
00:00:14.760 --> 00:00:18.040
I'm the founder of Atlan.

s5
00:00:18.040 --> 00:00:24.160
And today I'm going to talk about this thing where context is having its moment.

s6
00:00:24.160 --> 00:00:30.240
And so my goal today is to talk about like WTF is the context layer?

s7
00:00:30.240 --> 00:00:35.200
Just before I start and I promise this is the last time.

s8
00:00:35.720 --> 00:00:41.000
Um I don't know if the clicker is working.

s9
00:00:57.400 --> 00:01:00.080
Atlan, we it's it's working.

s10
00:01:00.080 --> 00:01:01.240
Thank you.

s11
00:01:01.240 --> 00:01:05.160
Um the problem we solve is we say AI doesn't know your business, we fix that.

s12
00:01:05.160 --> 00:01:15.480
We work with an incredible group of companies around the world ranging from GitLab and Zoom and Discord and Affirm to large enterprises like MasterCard and General Motors.

s13
00:01:15.480 --> 00:01:29.000
Um And about a year ago my co-founder and I went on stage and we said uh at the dawn of the internet era Bill Gates had written this very famous blog post and it said content

s14
00:01:29.000 --> 00:01:30.000
is king.

s15
00:01:30.000 --> 00:01:35.960
Um and as we are the dawn of the agentic era, context will be king.

s16
00:01:35.960 --> 00:01:39.440
Since then, it feels like 2026 is the year of context.

s17
00:01:39.440 --> 00:01:42.320
Context graphs, anyone?

s18
00:01:42.320 --> 00:01:47.320
Uh you know, every every two days you see some version of context popping up.

s19
00:01:47.320 --> 00:01:49.680
And so what is going on?

s20
00:01:49.680 --> 00:01:57.720
Um I believe the answer to this kind of is in this reality distortion field that we live in.

s21
00:01:57.720 --> 00:01:59.480
I live here in the Bay Area.

s22
00:01:59.480 --> 00:02:05.120
Every day or two I have conversations with people which kind of go like, "How far are we from AGI?"

s23
00:02:05.120 --> 00:02:09.000
and we have a debate and we're like, "Well, 1 year, 3 years, so on."

s24
00:02:09.000 --> 00:02:13.440
There is no doubt that the models are getting exponentially smarter by the day.

s25
00:02:13.440 --> 00:02:15.000
2 years ago they couldn't pass the bar.

s26
00:02:15.000 --> 00:02:19.360
Today, if they were to take the bar, it was they're the top 1% of test scorers.

s27
00:02:19.360 --> 00:02:24.560
On the other hand, they're not exponentially more useful by any benchmark.

s28
00:02:24.560 --> 00:02:30.080
1 out of 5, you know, AI use cases actually make it to production.

s29
00:02:30.080 --> 00:02:36.120
You know, 56% of CEOs say that there's zero financial benefit from AI today.

s30
00:02:36.959 --> 00:02:38.360
So, what's going on?

s31
00:02:38.360 --> 00:02:44.520
I believe hidden in plain sight is actually how performance is measured in the human world.

s32
00:02:44.520 --> 00:02:49.760
Cognitive intelligence doesn't really determine real-world effectiveness.

s33
00:02:49.760 --> 00:02:54.520
In fact, only 10% of job performance variance is explained by IQ.

s34
00:02:54.520 --> 00:02:55.360
Like, just think about it.

s35
00:02:55.360 --> 00:03:02.560
Would you say your smartest, you know, teammate who scored the highest on the SATs is also your best teammate?

s36
00:03:02.560 --> 00:03:10.080
Or would you say, "No, it's the person who works the most and takes the most feedback and learns the fastest?"

s37
00:03:11.360 --> 00:03:17.600
In the real world, we care about performance and performance is outcomes that you deliver in the real world.

s38
00:03:17.600 --> 00:03:19.680
And performance is a function of two things.

s39
00:03:19.680 --> 00:03:22.280
It's a function of intelligence, which is cognitive horsepower.

s40
00:03:22.280 --> 00:03:25.080
That's what the model benchmarks measure every day.

s41
00:03:25.080 --> 00:03:27.080
But, it's also a function of context.

s42
00:03:27.080 --> 00:03:30.880
This is what they say in the human world as learning on the job, right?

s43
00:03:30.880 --> 00:03:35.880
Knowledge and skills and expertise that you learn over time.

s44
00:03:35.880 --> 00:03:40.959
And in the last decade, we have compounded on one of those parameters.

s45
00:03:40.959 --> 00:03:44.280
Intelligence has 1,000x'd in the last decade.

s46
00:03:44.280 --> 00:03:48.120
Just in the last 6 months, we have 2x'd on that axis.

s47
00:03:48.120 --> 00:03:53.800
On the other hand, context, the situated knowledge of your business, that's barely moved.

s48
00:03:53.800 --> 00:03:57.520
We've moved some data to the cloud, uh but that's about it.

s49
00:03:57.520 --> 00:04:05.680
It's otherwise logged in dashboards and Slack threads and uh the head of that analyst who might be leaving next week.

s50
00:04:05.680 --> 00:04:16.320
Um And so, the question ahead of us then, I really believe this is the next frontier, is how do we help AI build context about our business?

s51
00:04:16.320 --> 00:04:25.720
Um And every time I'm faced with a question about how do we help AI do this, I always like to go back and understand how did we help humans do this?

s52
00:04:25.720 --> 00:04:32.440
Uh I'm going to take you into the life of, you know, uh uh exemplar employee, Maya.

s53
00:04:32.440 --> 00:04:39.440
Uh let's say she's a data analyst at Mac Context Burgers, cuz I thought I was going to be creative and I'm not very creative.

s54
00:04:39.440 --> 00:04:46.120
Um And, you know, let's say she's that analyst that everybody, you know, pings in your company.

s55
00:04:46.120 --> 00:04:46.680
Uh right?

s56
00:04:46.680 --> 00:04:50.720
She's the person that everybody sends a message to every morning when they're trying to solve a problem.

s57
00:04:50.720 --> 00:04:58.400
So, let's say this morning, uh there's a franchisee owner who sends her a message and says, "Why is my drive-thru time up this week?

s58
00:04:58.400 --> 00:05:00.600
Why is this metric up this week?"

s59
00:05:00.600 --> 00:05:03.080
Sounds like a really simple question.

s60
00:05:03.080 --> 00:05:07.120
Um But, it's actually a really complicated question to ask.

s61
00:05:07.120 --> 00:05:15.080
Just to answer this one very simple question, Maya first needs to know, uh what is drive-thru time?

s62
00:05:15.080 --> 00:05:16.120
Uh and who's asking?

s63
00:05:16.120 --> 00:05:19.600
Is it finance or is it, you know, my ops team?

s64
00:05:19.600 --> 00:05:21.520
And it might mean different things.

s65
00:05:21.520 --> 00:05:23.200
Uh But, not just that.

s66
00:05:23.200 --> 00:05:25.000
What does this week mean?

s67
00:05:25.000 --> 00:05:27.320
Is the cut-off period Monday to Sunday?

s68
00:05:27.320 --> 00:05:28.400
Is it Pacific time?

s69
00:05:28.400 --> 00:05:30.280
Is it Eastern time?

s70
00:05:30.280 --> 00:05:31.800
Uh That's knowledge.

s71
00:05:31.800 --> 00:05:32.640
Like, that's facts.

s72
00:05:32.640 --> 00:05:34.720
That's the map of the business.

s73
00:05:34.720 --> 00:05:36.440
Um But, not just that.

s74
00:05:36.440 --> 00:05:38.520
Uh there's expertise.

s75
00:05:38.520 --> 00:05:38.919
Uh right?

s76
00:05:38.919 --> 00:05:41.440
There's um you know, a diagnostic playbook.

s77
00:05:41.440 --> 00:05:43.360
What What does a great analyst do?

s78
00:05:43.360 --> 00:05:53.960
They know that, you know, quarter three is a season seasonal quarter because of weather patterns and they know to go check if the reason there's a spike is because of seasonality.

s79
00:05:53.960 --> 00:06:03.600
They also know that the company launched a product just that previous quarter and so they know to check if that's why the root cause analysis failed.

s80
00:06:03.600 --> 00:06:08.840
This is expertise and skills that people pick up over time as they learn on the job.

s81
00:06:08.840 --> 00:06:10.920
And then there's norms, right?

s82
00:06:10.920 --> 00:06:16.360
There's, you know, persona scoping, who's asking the question, how do I answer this question?

s83
00:06:16.360 --> 00:06:27.560
And Maya, she's one of those like cool people, she nails it, she sends an answer not just with the answer but with the why and the root cause and she finds the reason for it.

s84
00:06:27.720 --> 00:06:29.960
How did Maya learn to do this?

s85
00:06:29.960 --> 00:06:33.360
Um she just joined the company a year ago.

s86
00:06:33.360 --> 00:06:41.640
First, Maya, you know, has for like she joined and she got some training like all of us do, but that's not where any of us learn, right?

s87
00:06:41.640 --> 00:06:42.720
In our companies.

s88
00:06:42.720 --> 00:06:43.800
How do we learn?

s89
00:06:43.800 --> 00:06:52.440
We learn because you shadow like the best teammate and then you see why they're doing something and then you learn from that and then you make a mistake.

s90
00:06:52.440 --> 00:06:57.640
Who here has learned more from a mistake than anything else?

s91
00:06:57.640 --> 00:06:58.440
Right?

s92
00:06:58.440 --> 00:07:01.480
You make a mistake and then you learn.

s93
00:07:01.480 --> 00:07:05.120
You your manager gives you feedback and you learn not to do that again.

s94
00:07:05.120 --> 00:07:08.760
You deal with an edge case and then you learn from that.

s95
00:07:08.760 --> 00:07:12.640
That's how all of us humans learn at work.

s96
00:07:12.640 --> 00:07:18.280
And so then the question is how do you help build the agentic Maya?

s97
00:07:18.280 --> 00:07:23.840
And now I want to walk you through our experiments and learnings as we've built this at Atlan.

s98
00:07:23.840 --> 00:07:28.520
Era one and this was roughly about 18 months ago now.

s99
00:07:28.520 --> 00:07:34.200
We started on the the track of bootstrapping agents.

s100
00:07:34.200 --> 00:07:42.360
And the way we went about it was and we started this with our customer experience team and we did this jobs to be done analysis map, right?

s101
00:07:42.360 --> 00:07:49.480
And so we said, "Hey, if you are someone on our customer experience team, what are all the things that you do on a day-to-day basis?"

s102
00:07:49.480 --> 00:07:51.120
And then we made some hypothesis.

s103
00:07:51.120 --> 00:07:56.400
We We said, you know, for example, one part of the job is documentation and meeting prep.

s104
00:07:56.400 --> 00:08:00.640
Uh we said, "Well, AI could probably do that job pretty well."

s105
00:08:00.640 --> 00:08:02.200
Uh and so we built the scaling factor.

s106
00:08:02.200 --> 00:08:06.760
So, on the other hand, relationship management is something that our customer experience team does.

s107
00:08:06.760 --> 00:08:10.920
And we said, "Hmm, that doesn't sound like something AI is going to be able to do anytime soon."

s108
00:08:10.920 --> 00:08:13.220
And so we built the scaling factor.

s109
00:08:13.220 --> 00:08:13.400
[snorts]

s110
00:08:13.400 --> 00:08:23.280
And then we basically started bootstrapping these individual agents that were like built for that specific topic.

s111
00:08:23.280 --> 00:08:34.760
Uh our team got creative, so we had Hermione, who is our health intelligence lead, and then we had, you know, MoneyPenny, who was our financial risk analyst, and we just made that particular agent

s112
00:08:34.760 --> 00:08:38.159
really good at doing that one thing.

s113
00:08:38.159 --> 00:08:41.240
Um and that worked for some time.

s114
00:08:41.240 --> 00:08:46.200
Um but then we realized there were some challenges with this approach.

s115
00:08:46.200 --> 00:08:48.880
The first, context engineering.

s116
00:08:48.880 --> 00:08:52.560
Uh we got to the point by middle of last year where building an agent was really easy.

s117
00:08:52.560 --> 00:08:54.080
Took like 5 minutes.

s118
00:08:54.080 --> 00:09:00.240
Uh but giving it the business context that it took to actually get it to be accurate took forever.

s119
00:09:00.240 --> 00:09:10.861
Uh quality of the agent often dependent uh on the quality of context engineering, and that led to a lot of weird lost trust cases with our stakeholders.

s120
00:09:10.861 --> 00:09:11.120
[snorts]

s121
00:09:11.120 --> 00:09:20.560
Um then, as we started taking this into production, we started seeing that these agents basically were kind of like living on their own island.

s122
00:09:20.560 --> 00:09:30.080
Uh now imagine, for example, if you're in a human team, and your marketing changes positioning on your, you know, and then they come to the town hall and they tell you that they changed positioning.

s123
00:09:30.080 --> 00:09:36.160
And so then, you know, the SDR on your team or your sales development rep, they know that they should use is new positioning.

s124
00:09:36.160 --> 00:09:40.840
This is like the infrastructure that we've built for humans inside our organizations.

s125
00:09:40.840 --> 00:09:42.520
Agents didn't have them.

s126
00:09:42.520 --> 00:09:51.520
So, our marketing team had these agents and they started making changes to that and then our SDR agent on our website was still pitching the old version.

s127
00:09:51.520 --> 00:09:54.280
Uh we had no idea how any of these things were even connected.

s128
00:09:54.280 --> 00:09:58.800
So, we didn't even know how to like run this as as a team of agents.

s129
00:09:58.800 --> 00:10:02.360
Uh when an agent gets something wrong, this is hard.

s130
00:10:02.360 --> 00:10:04.840
Uh it was really hard to like trace back what happened.

s131
00:10:04.840 --> 00:10:06.080
Was it the model?

s132
00:10:06.080 --> 00:10:06.560
The agent?

s133
00:10:06.560 --> 00:10:07.480
Was it the context?

s134
00:10:07.480 --> 00:10:10.280
Like where How do we even go back and fix this?

s135
00:10:10.280 --> 00:10:15.840
Um and over time we started dealing with context sprawl.

s136
00:10:15.840 --> 00:10:23.000
Uh we had the The hard part about this was agents all had their own memory systems to a certain extent.

s137
00:10:23.000 --> 00:10:27.400
So, they were learning They were all learning separately and they were learning differently.

s138
00:10:27.400 --> 00:10:34.160
Uh it became very, very difficult very quickly to say, "Okay, what does the single version of truth here look like?"

s139
00:10:34.160 --> 00:10:41.520
Um and then over time we actually went through In the last 12 months we've gone through cycles of at the agent tech layer.

s140
00:10:41.520 --> 00:10:46.880
About 12 months ago we were using one of these no-code type builders uh called Relevance.

s141
00:10:46.880 --> 00:10:50.600
We went from there into Google ADK, then we tried Glean.

s142
00:10:50.600 --> 00:10:53.200
Uh start of this year we moved to Claude Code.

s143
00:10:53.200 --> 00:10:56.839
Now we are kind of like 50/50 Claude and Codex.

s144
00:10:56.839 --> 00:11:04.600
Um and every single time as these changes happened, uh our context got trapped in each of these individual systems.

s145
00:11:04.600 --> 00:11:15.760
Um So, starting this year as general purpose agents started to become a thing, we said, "What if there was a different approach with general purpose agents?"

s146
00:11:15.760 --> 00:11:20.320
Uh again, going back to the human world, well, Maya, she's not an individual star.

s147
00:11:20.320 --> 00:11:21.720
She's part of a team, right?

s148
00:11:21.720 --> 00:11:28.560
And you know, you talk about these dream teams like Maya and someone who runs customer support and someone who launches ads.

s149
00:11:28.560 --> 00:11:30.600
These people work really well together.

s150
00:11:30.600 --> 00:11:35.200
And often these dream teams are built on shared context, right?

s151
00:11:35.200 --> 00:11:37.640
They have a shared language.

s152
00:11:37.640 --> 00:11:40.560
They have a shared picture of what's true today.

s153
00:11:40.560 --> 00:11:43.040
They have shared playbooks.

s154
00:11:43.040 --> 00:11:46.840
They have shared norms, who's allowed to make what decision.

s155
00:11:46.840 --> 00:11:48.160
And then they learn together.

s156
00:11:48.160 --> 00:11:49.960
I think this is the most important part of it.

s157
00:11:49.960 --> 00:11:53.560
They have compounding learning loops of what good looks like.

s158
00:11:53.560 --> 00:12:01.600
And they have shared memory that, you know, oh, we launched this thing last quarter and it like was terrible and we're not going to make that mistake again, right?

s159
00:12:01.600 --> 00:12:07.560
And so we said, is there a way to bring that into the way we think about AI in our companies?

s160
00:12:07.560 --> 00:12:19.560
And so the mental model we started working on was we said, okay, we have these teams of humans and they're across the board and can these people essentially start building domain skills.

s161
00:12:19.560 --> 00:12:23.840
So each of them is responsible for a certain set of skills.

s162
00:12:23.840 --> 00:12:30.480
All of this goes into this common one place, which is this one company brain of sorts, right?

s163
00:12:30.480 --> 00:12:32.960
I like to think of this as the context layer.

s164
00:12:32.960 --> 00:12:40.640
And then this has a bunch of retrieval mechanisms, which then talks to the general purpose agent across the ecosystem.

s165
00:12:41.400 --> 00:12:44.200
So then we started the experiment.

s166
00:12:44.200 --> 00:12:47.839
This is some version of what our marketing team ended up building.

s167
00:12:47.839 --> 00:12:51.920
So you'll see on the left, those are all the systems that our marketing team uses.

s168
00:12:51.920 --> 00:12:58.760
So data systems, our social and community platforms, our ad platforms, our analytics platforms.

s169
00:12:58.760 --> 00:13:02.160
And then you'll see this agent block.

s170
00:13:02.160 --> 00:13:07.960
We built this very specifically for having openness.

s171
00:13:07.960 --> 00:13:07.980
So

s172
00:13:07.980 --> 00:13:08.240
[snorts]

s173
00:13:08.240 --> 00:13:10.120
we had Cloud Code and Co-work.

s174
00:13:10.120 --> 00:13:17.240
We also had our own claw that we deployed, which has, you know, essentially talks in our Slack channels.

s175
00:13:17.240 --> 00:13:21.074
And then we used some external products like Qualified and Artisan.

s176
00:13:21.074 --> 00:13:22.760
[snorts]

s177
00:13:22.760 --> 00:13:26.200
In the middle is kind of this context layer that our team started building.

s178
00:13:26.200 --> 00:13:32.120
So think of it as our best SEO person was building the SEO skill.

s179
00:13:32.120 --> 00:13:36.560
Uh our best competitive intel person was building the best competitive intel skill.

s180
00:13:36.560 --> 00:13:42.034
And that kind of became this common repo that we were building into and pulling out from.

s181
00:13:42.034 --> 00:13:43.920
[snorts]

s182
00:13:43.920 --> 00:13:46.280
This sort of became our living brain.

s183
00:13:46.280 --> 00:13:49.960
Over time, we realized there were some things that we needed in this brain, right?

s184
00:13:49.960 --> 00:13:51.920
Uh we realized we needed a data graph.

s185
00:13:51.920 --> 00:13:57.800
Like if For example, our autonomous ads agent we realized it needs to do analysis on a daily basis.

s186
00:13:57.800 --> 00:14:00.520
So, like which table should I go pull from?

s187
00:14:00.520 --> 00:14:02.840
Uh we needed a library of skills.

s188
00:14:02.840 --> 00:14:16.040
We also needed some other things, semantics, metrics, what is ARR, how do you measure that, uh what is a qualified lead in our company, uh and org structure, entities, things like that.

s189
00:14:16.120 --> 00:14:24.000
Over the last 6 months, we ended up creating about 300 skills and 40 agents in this team, uh which has been incredible.

s190
00:14:24.000 --> 00:14:28.920
Uh but then, with this approach, too, we realized that there were some challenges.

s191
00:14:28.920 --> 00:14:34.320
We realized that context kind of needs to be managed like code.

s192
00:14:34.320 --> 00:14:37.040
Um so, some challenges.

s193
00:14:37.040 --> 00:14:38.560
Let's pick skills.

s194
00:14:38.560 --> 00:14:41.800
Uh dependency management became really complicated.

s195
00:14:41.800 --> 00:14:49.400
So, for example, we have this competitive intelligence skill, and it learns from the market on what's changing in the market, and it improves.

s196
00:14:49.400 --> 00:14:57.080
Um it feeds our category positioning skill, which then feeds our sales battle card skill.

s197
00:14:57.080 --> 00:15:04.000
Uh now, each of these skills is learning and evolving, uh but every time they learn and evolve, it breaks something downstream.

s198
00:15:04.000 --> 00:15:08.200
Uh and these skills very quickly start getting outdated and start drifting.

s199
00:15:08.200 --> 00:15:10.520
Uh who owns skill quality became another thing.

s200
00:15:10.520 --> 00:15:14.040
Like who eventually owns the quality of this?

s201
00:15:14.040 --> 00:15:16.080
Security and governance was a nightmare.

s202
00:15:16.080 --> 00:15:19.560
Uh we had secrets hardcoded in .env files.

s203
00:15:19.560 --> 00:15:25.320
Uh it was People were downloading these public skill repos, this the whole thing was like a nightmare.

s204
00:15:25.320 --> 00:15:30.360
And then I talked about context portability across all these multi-agent systems.

s205
00:15:31.600 --> 00:15:35.840
I started this talk by saying WTF is a context layer.

s206
00:15:35.840 --> 00:15:39.800
These are the problems that a context layer is meant to solve.

s207
00:15:39.800 --> 00:15:45.720
The question I like to ask is what does the GitHub for context look like?

s208
00:15:45.760 --> 00:15:47.800
Few thoughts.

s209
00:15:47.800 --> 00:15:54.960
Company context needs life cycle management and collaboration and versioning just like code does.

s210
00:15:56.160 --> 00:16:02.520
You know, there's questions like what's local context, what's global context, how do I keep this updated, so on.

s211
00:16:02.520 --> 00:16:08.920
Some thoughts in this, can skills have a profile just like code does?

s212
00:16:09.080 --> 00:16:13.720
Can that have a self-learning loop that's baked into it?

s213
00:16:13.720 --> 00:16:15.760
What does quality management look like?

s214
00:16:15.760 --> 00:16:20.080
Can you have security and postures posture management associated with that?

s215
00:16:20.080 --> 00:16:23.000
That's really like the first step.

s216
00:16:23.240 --> 00:16:28.080
I see this as like having something that has built-in versioning and quality and dependency management.

s217
00:16:28.080 --> 00:16:31.560
So, you should be able to say, "Hey, this thing impacts all these other things.

s218
00:16:31.560 --> 00:16:35.120
This is the approver, this is the maintainer, these are the contributors.

s219
00:16:35.120 --> 00:16:41.506
How do you build that kind of human plus AI workspaces that these that these skills are managed via?"

s220
00:16:41.506 --> 00:16:43.000
[snorts]

s221
00:16:43.000 --> 00:16:50.040
Second thing, every AI interaction creates more context and harnessing this is gold.

s222
00:16:50.040 --> 00:16:53.240
There's been I know a lot of talks about self-improving loops.

s223
00:16:53.240 --> 00:17:02.480
We have found that with traces deploying a specific harness that actually is specialized in being able to go and reverse construct from that.

s224
00:17:02.480 --> 00:17:10.520
So, think of it as AI that's reading through all your traces and almost brings it back to your maintainer loop and says, "Approve, reject.

s225
00:17:10.520 --> 00:17:11.360
Approve, reject.

s226
00:17:11.360 --> 00:17:13.079
Improve this over time."

s227
00:17:13.079 --> 00:17:16.319
That's the compounding learning loop.

s228
00:17:16.480 --> 00:17:21.560
And the third, often a lot of people ask me this question, which is like, how do I start?

s229
00:17:21.560 --> 00:17:26.959
Because my business is like really disparate, and I have all these like 60 systems, and how do I even start?

s230
00:17:26.959 --> 00:17:31.840
One of the biggest learnings we've had is context is hidden in these in business systems.

s231
00:17:31.840 --> 00:17:36.160
Uh and across this context quality can really compound.

s232
00:17:36.160 --> 00:17:47.400
So, for example, if you're able to connect your Salesforce and your HubSpot to your data warehouse, to your application layer, and then you're able to reverse construct how these things are actually connected one to another,

s233
00:17:47.400 --> 00:17:50.280
context today gets lost in every one of those hops.

s234
00:17:50.280 --> 00:18:02.727
But if you can reverse construct that, and then deploy AI on top of it, we've seen incredible accuracy in being able to reverse construct the first version of your company brain.

s235
00:18:02.727 --> 00:18:04.727
[snorts]

s236
00:18:05.520 --> 00:18:07.120
So, I'll end with this.

s237
00:18:07.120 --> 00:18:18.600
Uh the way I think about a context layer is it's a system that turns knowledge and expertise and norms that we talked about, that Maya knows, into a machine usable context

s238
00:18:18.600 --> 00:18:20.200
for AI systems.

s239
00:18:20.200 --> 00:18:27.240
Uh At a very high level, the way I like to think of it is it looks like this.

s240
00:18:27.240 --> 00:18:32.160
Uh It continually is mining context from your business systems.

s241
00:18:32.160 --> 00:18:35.560
It's feeding this in to that one company brain.

s242
00:18:35.560 --> 00:18:42.840
It's harnessing this in skills and context development life cycles as your teams go and deploy these agents.

s243
00:18:42.840 --> 00:18:45.600
And then it has a bunch of ways you can retrieve it.

s244
00:18:45.600 --> 00:18:55.783
So, MCP, SQL, vector retrieval, hybrid assembly, all these different ways that you retrieve it, and pull back from traces and build this compounding learning loop.

s245
00:18:55.783 --> 00:18:57.783
[snorts]

s246
00:18:58.440 --> 00:19:04.840
Today, we're largely building agents by hardcoding context.

s247
00:19:07.240 --> 00:19:13.120
The scale of this problem I truly believe is underhyped, because with scale this can become really unsustainable.

s248
00:19:13.120 --> 00:19:15.400
Uh And a little dangerous.

s249
00:19:15.400 --> 00:19:23.360
Like all of us know this this old joke, which is if you ask sales and finance the revenue number, you're going to get two different numbers.

s250
00:19:23.360 --> 00:19:30.720
We're fast approaching a moment of starting to deploy autonomous systems where the same thing is starting to happen.

s251
00:19:31.400 --> 00:19:32.840
So, I'll end with one last thing.

s252
00:19:32.840 --> 00:19:37.280
I started this presentation by saying context is king.

s253
00:19:37.520 --> 00:19:42.320
I'd like to end it by saying context is also IP.

s254
00:19:42.320 --> 00:19:52.760
Something I think a lot about is in a world where you and your competitor have access to the same models and the same intelligence, what differentiates a company?

s255
00:19:52.760 --> 00:19:59.720
What differentiates a customer support agent at American Express versus Amazon?

s256
00:19:59.720 --> 00:20:01.960
That's how you do business.

s257
00:20:01.960 --> 00:20:04.840
That's what makes your company special.

s258
00:20:04.840 --> 00:20:15.960
Context is how we take and encode our culture and our norms into something that we will be proud of as we build autonomous frontier firms.

s259
00:20:15.960 --> 00:20:19.000
Um And that's all I had.

s260
00:20:19.000 --> 00:20:24.320
Um you can find me at Prokalpa on Twitter um or write to me.

s261
00:20:24.320 --> 00:20:30.600
We are actively working with folks on the frontier on going and shipping and building company brains.

s262
00:20:30.600 --> 00:20:33.320
Um so, if you'd like to talk to us, feel free to reach out.

s263
00:20:33.320 --> 00:20:35.344
Thank you.

s264
00:20:35.344 --> 00:20:37.344
[applause]

s265
00:20:49.754 --> 00:20:51.754
[music]