AI Writes the Code. So What's Left for Engineers? - Laurie Voss, Arize AI

AI Engineer · 27 min · 331 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.

00:01[music]
00:12My name is Laurie.
00:13I'm head of developer relations at Rise AI.
00:16Uh in a former life I co-founded NPM Inc. So some of may you some of you may remember me from JavaScript days.
00:21Uh but these days I think about AI agents and uh in this case what they are doing to our jobs.
00:28Uh so a little bit over 3 years ago um uh Swix wrote an essay called the rise of the AI engineer and his argument was that when it comes to shipping AI products you want engineers not researchers.
00:41I think that is a message that resonated with everybody especially who the people who come to this conference.
00:47Uh and so he created the job title of the AI engineer and companies since then have been hiring for it.
00:53Um and part of what I'm here to do today to do is to convince you uh that he did not in fact go far enough.
01:01Uh there's not just an AI engineer revolution happening.
01:05There is a revolution happening to every kind of engineer.
01:09Um all engineering jobs are being replaced by a new type of job and not everybody is going to be happy to get that news.
01:17Um the underlying change that kicked all of this off is that agents have made writing code basically free.
01:25Like you pay for the tokens but relative to the cost of the time it used to take you uh it is basically free.
01:31Um everything that comes after writing the code did not become free at the same time.
01:36So the the bottleneck moved from writing code is really slow and all of this other stuff happens at roughly the same speed to writing code is super fast uh and now we get stuck behind new things.
01:46So the first the thing it's stuck behind right this second like this month is code review.
01:51Um uh but my thesis is that it's not going to stop there.
01:57We are going to figure out how to automate code review.
01:59We are going to make uh code review move at the speed of agents, and then we are going to run into another bottleneck, and another bottleneck, and another bottleneck, and at some point we're going to stop.
02:10We're going to hit one that we can't get past.
02:12Uh and that is what is going to be left for the humans to do.
02:17Whatever the bottleneck is that we can't get agents past.
02:20So, I'm making a bet that I can think of a thing that agents really are not going to be able to do.
02:27Uh and then once we've come to that bottleneck, we're going to have to figure out how do we train the next generation of engineers to be able to do that job because we're not training them to do that job right now.
02:39So, a quick warning {slash} apology.
02:40I have 18 minutes because that is how long Swiss gives for talks.
02:44Uh and I have a really big pile of data.
02:47I did a ton of research for this.
02:49Uh and I don't have time to give it all to you.
02:50I don't have time to like quote my sources and do all that stuff.
02:53So, I'm just going to throw numbers at you with a bunch of citations on the slides, and I'm just going to have to sort of ask you to trust me, bro.
03:01Uh
03:02[laughter]
03:03Uh so, developers who switch on autonomous coding agents write 741% more code, eight times as much.
03:11Uh and but they ship only 30% faster.
03:14Like, 30% faster is not nothing.
03:1530% faster is amazing.
03:17We are shipping a lot more code than we used to, but given that we're writing eight times more code, only a third more actual shipping is kind of bad, right?
03:26It means that we're stuck somewhere.
03:28Uh and where we're stuck is code review.
03:30Um and the reason that we're stuck is because humans can't just review faster.
03:35Uh there is an enormous study done by Cisco 20 years ago about how fast you can do code review, and how accurately you can do code review.
03:45Um and it turns out that humans' ability to like parse what's going on in code falls off radically after about 450 lines of code per hour.
03:56So, if you've produced a 10,000 line diff with your agent, it would take you like a week to reasonably review that code.
04:04And the answer is that nobody's actually doing that, right?
04:08Cuz we're all throwing 10,000 diffs around willy-nilly these days.
04:12Um so, I did a whole talk about this phenomenon at a previous AIE, and that talk is called the death of the code review.
04:18And if you want all of the information about the death of the code review, you can watch that talk.
04:22Um but the problem facing us right now, uh which is not the same as the problem I'm trying to solve with this overall with this talk, is that the cost of producing
04:31plausible code has collapsed while the cost of knowing whether to trust that code has stayed exactly the same.
04:38Um so, the obvious review move right now is to make the machines do the review, like I said.
04:43And that is already happening.
04:45Um GitHub's Copilot uh does more than one in five of the code reviews on GitHub as of last month.
04:52Uh you can, if you want to, opt in to let uh Copilot be the final reviewer before something goes to uh production.
05:00You can make it have the approval power, if you want to, which is crazy.
05:05I don't know who's turning on that option, but you can do it if you want to.
05:08Um OpenAI says that at at OpenAI, humans may review their pull requests, but aren't required to.
05:15Um but the problem is that the machine reviewers aren't very good yet.
05:19Uh the data says that there's uh the data that I found put four AI reviewers through their paces, and they all found a bunch of bugs, but none of the same bugs.
05:30They did not agree on any individual bug.
05:33Uh they all found actual issues in the code.
05:36They all found different issues in the code.
05:38Um so, they're worth running, but if you are serious about this, you're never going to run just one of them.
05:43Uh because they are not reliable enough yet.
05:47Um And there are some people who went all the way.
05:51They were like, "I'm just not going to review the code anymore.
05:53I'm just going to assume my agent has one shot at this perfectly."
05:57Um Dexter Horty, who is here, I don't know if he's in the audience, probably not.
06:01Um uh he spent 6 months telling people not to read the code, and then after that 6 months, he got on stage and said, "Nope, I was wrong.
06:09You can't actually uh skip code review entirely.
06:12We tried that.
06:13It was a disaster.
06:14We had to rip replace whole systems."
06:16Uh that's not an option.
06:19The agents are not good enough.
06:20They're not one shotting the code.
06:22Um Amazon had a run of outages uh that they did postmortems on and decided that it was because their coding agent had produced a bunch of slop and they just released it into the wild without thinking.
06:33Uh and so now they don't let them do that anymore.
06:36Uh senior engineers must sign off on code reviews at Amazon.
06:39Um and Anthropic published a postmortem uh on bugs that had got past every reviewer that they had, both human and machine.
06:46So, the problem is that the automations that we've built so far to check whether our code is good enough are not themselves good enough.
06:53Um Claude Fable 5 passes the tests uh on 80% of a standard coding benchmark.
06:59Um and co- So, Cognition, the makers of Devin, um they built a benchmark that asks a different question, which is not does it pass the tests?
07:08The question is would you as the reviewer, as the maintainer of this repo, merge this PR?
07:14And if you ask that question instead, the pass rate falls from 87% uh to 54%.
07:21Um because there's a lot of extra information that goes into deciding whether you would merge it.
07:28There's code style, there's side effects, there's external systems, there's history that is not encoded into the code base.
07:35It is encoded into the brain of the maintainer.
07:38Um So, what the maintainer is adding, what maintainer is demonstrably adding, is something that we haven't given the agents yet.
07:47There is information they don't have, there is a skill they don't have.
07:51Um and it is the sense of what good means.
07:55That is what we are That is what the agents don't have yet.
07:58Uh and that is a critical insight that I'm going to come back to.
08:01Um Eliyahu Goldratt, if I'm pronouncing that correctly, was a physicist who became a management guru and in 1984 he wrote a a novel about running a factory.
08:11Does that sound like an interesting novel to you?
08:14Um it is called The Goal.
08:16Um and his big idea was the theory of constraints.
08:19Uh every system has one bottleneck and speeding things up does not work unless the thing that you were speeding up is the bottleneck.
08:26Uh that is our problem with code review.
08:30Uh we've sped up the code and we have not sped up the review.
08:33Um his second idea is the one that matters today, which is that when you fix the bottleneck, it doesn't go away, it just moves and finds another bottleneck to pile up against.
08:42Um so let's assume, despite everything that I've said about how bad the code reviewers are right now, that we figured that out, that we get automated code review to a higher standard
08:53and we move that bottleneck.
08:55What is the next bottleneck uh that it is going to run into?
08:59Um if we figure out review, the next bottleneck is probably going to be operations.
09:04How do I deploy this thing?
09:05How do I scale it up?
09:07Uh and then you're probably going to run into security.
09:10How do I make this thing safe at scale?
09:13Uh how do I make sure that this thing is private?
09:15How do I make sure that it's not radically going off the rails?
09:18Uh but I'm going to make and this is the big leap that I'm making in this talk.
09:23I'm going to assume we figured those things out as well because those things are very heavily solved by humans.
09:32There's an enormous amount of data out there about them.
09:34If we can figure out how to do operations, how to do scaling, how to do security using progressively more advanced agents, what is it that we are left to do?
09:46Like, what is the role of a human in that situation?
09:49Um it's also it's a big question [clears throat] and it's also a long-term question.
09:54I'm not telling you what is going to happen in the next 12 months.
09:56In the next 12 months, I think we'll probably make some real progress on code reviews.
10:00I do not think 12 months from now we'll have sorted out operations and scaling and security.
10:05I think that is going to take somewhere between 5 and 10 years to get done.
10:09Uh despite people who are like, "AGI 2027."
10:12No, we're going to slowly iterate on this stuff.
10:16Um But the other assumption that you need to make for my for the my thesis to be correct is that the amount of software that we need is effectively infinite.
10:26And I feel that one is a much easier case to make.
10:29Think of every website you've ever run into uh in the world in production, and it's probably pretty bad.
10:36Uh and the reason it's bad is not because we don't know how to make really great websites.
10:41The problem is that there's not enough people to make all of the websites good.
10:45Every time you run into a terrible website at your hairdresser or the government or like, you know, your bank, like it's not because there's not somebody who knows how to make that website good.
10:53It's because A, it's very expensive to make it good, and B, the people who are doing that are busy doing other things.
11:00Uh you know, probably shipping AI agents.
11:03Um so assume that we can make as much software as we want basically for free, and that the appetite for making software is essentially infinite.
11:13That is a wonderful situation in which to be in.
11:15We've got infinite demand and also suddenly infinite supply.
11:19Uh where does the bottleneck end up?
11:21What do we think the agents are not going to eat?
11:23Um to try and figure that out, I split up the job of software development into as many pieces as I could think of, and I sorted them by how soon the agents
11:31uh get to them.
11:32Uh writing code is already collapsed, reviewing and maintaining code is going soon.
11:37Um shipping to production and scaling are next on the chopping block, although there isn't a lot of data there yet.
11:42Uh and then there's the stuff that I think is probably safe, which is deciding what to build, uh defining what good means, and finally making it delightful.
11:54What's left is finding out what people actually want.
11:57Every piece of software is a formalization of something that a human wanted to get done.
12:03Uh So, if a customer tells you, "I need to to keep track of my orders," there are 10,000 pieces of software that keep track of your orders.
12:13That's what they do.
12:15Uh but only one of those, if you're lucky, is good if the thing that you're doing is you're running a bakery.
12:21If you're running like an auto parts store, that's going to be different.
12:24If you are running uh you know, a giant manufacturing concern, it's going to be completely different.
12:30The That requirement is not sufficient.
12:33The whole job is figuring out what exactly do you mean when you say that you want to keep track of your orders.
12:39What does that mean for your business?
12:41What does it have to do?
12:42How does that work?
12:44Uh and that knowledge isn't in the training data.
12:46It isn't in the weights.
12:47It is in the mind of the person who runs that business.
12:50And the only way to get it out is by flapping your meat parts at them and having them flap their meat parts back.
12:56Like, you have to talk to them.
12:58That is the only way to get the information out.
13:00Um And that work has a very important property, which is that it doesn't transfer.
13:06Uh if you think about what good means for like a calendar app and what good means for a scheduling app, those are solving almost the same problem, right?
13:13A calendar app and a scheduling [clears throat] app, but they're completely different pieces of software.
13:17They have totally different requirements and a totally different definition of good, even though [clears throat] those two use cases are so close together, and we have a zillion use cases.
13:26Um Anytime that you've used software that has a zillion configuration options and it still doesn't
13:32[snorts]
13:32do what you need.
13:33I'm looking at you, Jira.
13:35Uh like that is the problem you're running into of software that is trying to do everything and still failing at it.
13:43Um so the bottleneck moves to describing the problem and defining what it means to have solved the problem and my big bet for this talk and for the next 10 years is that is where it stays.
13:56Uh Fred Brooks, who ran IBM System 360 project and wrote uh The Mythical Man-Month, uh he wrote about this in 1986.
14:04He split the difficulty of software into two kinds.
14:06Accidental complexity is the difficulty of the tools or the languages, the typing, the syntax, and then there's the essential complexity, which is the prob- difficulty of the problem itself.
14:16Uh and he predicted that no single tool would make us 10 times more productive because the essential part is the hard part.
14:23Uh and Brooks's prediction is looking pretty solid because we made code eight times faster and we still didn't get very much faster at shipping because the hard part is the essential part.
14:34Um Shawn Grove from OpenAI made the same point uh 2 years ago.
14:38Uh no, sorry, a year ago.
14:40Uh what year is it?
14:41Um he said that code is a lossy projection of the specification and whoever writes the spec, if it's a PM or a lawmaker or an engineer or a marketer,
14:50they become the programmer.
14:52Uh and so writing [clears throat] down what good means is where the value is now.
15:00Uh I'm beginning to run out of time because I haven't timed this well.
15:06Um anything that you can verify cheaply, you can train against until you beat it if you are model maker.
15:14Uh so uh Cognition tried to write down because of their benchmark about what what it counts to merge something, does it, you know, is is this code mergeable?
15:23They were like, let's see if we can figure out a benchmark that defines what mergeable really means.
15:28Let's see if we can get all of that other information into the benchmark, and then we can train a model on it.
15:35Um uh Cursor's reviewer BugBot turns every thumbs down that you that your team gives it into a rule, and they have like 44,000 rules now.
15:45Uh so the moment the maintainer takes knowledge out of their brain and puts it somewhere written down that a model can train on it,
15:53[laughter]
15:53uh the bottleneck moves again.
15:55That is why I think security can go.
15:57That is why I think operations can go.
15:59I think we can write those things down eventually, even though we haven't done it yet.
16:03Uh Andrew Ng, the co-founder of Google Brain and Coursera Coursera, says that one of his team proposed having twice as many product managers as engineers, uh and he said that that was a first in his life, uh but his answer is since then hasn't been hire more PMs.
16:19It's to blur the jobs together.
16:21In his words, when an engineer understands users and can make decisions on what to build and build it directly, uh they can execute incredibly quickly, and that is where I think we are beginning to hit on the answer.
16:33Uh in the 1960s, there there was a job called the systems analyst.
16:38Their job was not to write code.
16:40Their job was to talk to the business and find out what the problem was.
16:44Uh then that analyst turned into something that we call a product manager today, and we've had two professions uh where there used to be one and a half.
16:52I think what is happening is they are collapsing back into one.
16:56We are going back to the '60s.
16:58So that is not an AI engineer.
17:01They're not just building AI.
17:02They're not just building with AI.
17:04They are building everything.
17:06They are building all software.
17:08Uh and I've decided to call this job the product engineer.
17:11This is somebody who finds out what people need, defines what good looks like, and ships it uh with agents doing the typing um Gergo Orosz whose name I have probably butchered he writes the pragmatic engineer newsletter he described
17:28the product-minded engineer back in 2019 as a nice trait for an engineer to have and what I'm saying now is it has gone from nice to essential.
17:36It has gone from the from a skill that senior engineers used to have to a skill that every engineer has to have if they're going to be an engineer.
17:47um And you can see the market already paying for this.
17:52Forward deployed engineers who look very much like product engineers job postings for them rose 800% in 9 months of 2025.
18:01They are everywhere everybody is trying to hire them.
18:03Who in this room is trying to hire a forward deployed engineer show of hands.
18:08Okay, well more I was expecting more hands oh well.
18:11uh So uh The average comp for forward deployed forward deployed engineer by the way is $240,000 which is to say that it pays like a senior engineer because it is a senior engineer.
18:23It's a very useful kind of engineer.
18:26um But nobody agrees on what to call it yet.
18:29There are like a thousand different job titles for this thing that I am describing and have decided arbitrarily to call the product engineer.
18:36um And probably at this point some of you are thinking that isn't programming and you are correct.
18:43It is not programming.
18:44The programming is being done by an agent.
18:46It is however recognizably still software development and we are going to have to wrap our minds around the idea that being a software developer and being a programmer have become divorced
19:00in 2026.
19:02um And when I say stuff like well I think the hard part is deciding what good is I realize that I'm falling into a trap that lots of people in AI land fall into, which is all of the parts everything about everything can be automated by AI except the thing that I'm good at because that's too hard.
19:20I heard a VC say that the other day and I was like, "Buddy, uh like, no. No, someone is going to automate your job any day now.
19:29Um uh so you have to take this with a grain of salt.
19:34Am I Am I just thinking that the thing that I enjoy about software development, which is figuring out what the problem is, is the hard part of software?
19:41You can tell me if I'm wrong.
19:43Um but say you believe me so far.
19:47Programming goes to zero, product engineering goes to infinity.
19:51That brings us to a new and important problem, which is junior developers because the way we used to train a junior developer is we used to give them very well-specified
20:01tickets that said, "Write this software this way in particular and then give it to me so that I can tell you all the ways that you were wrong."
20:09Uh and that is gone because that's what we do with the That's what we do with the coding agents now.
20:16We say, "Write this thing and then I'll tell you how you're wrong."
20:19And then we have a conversation with them.
20:21Uh and there's That is not a job that we need a human to do anymore.
20:25But we still need an enormous number of forward-deployed engineers or uh product engineers and we have to get them from somewhere.
20:34We can't just abandon the current generation of new developers who've showed up and said, "How do I do this stuff?"
20:39We're like, "I don't know because I don't know what your job is anymore."
20:44Uh
20:45[snorts]
20:45this is not a theoretical future.
20:47This is an actual graph of labor statistics since January 2023.
20:51It is sorted by age.
20:52The red line is engineers aged 22 to 25.
20:56Their employment is collapsing.
20:58We are not hiring junior engineers.
21:00This is not a thing that I think is going to happen.
21:02This is a thing that's been happening for 2 years.
21:04Job market sucks for them.
21:06Meanwhile, the job for older engineers 41 to 49 has gone way up because senior engineers are still super valuable, more valuable than before, but junior engineers are being screwed and we need to fix that as a profession.
21:20Uh so if we're going to reap the benefits of a world where we 10x code output and figured out the rest of the software supply chain and figured out what humans need to do,
21:29uh then we are going to need a lot of people to do that.
21:31I said that the demand for the software is going to infinity.
21:34We need infinity people.
21:35Where are we going to get them?
21:37Uh we've been creating them accidentally so far and we need to figure out a way to do them on purpose.
21:42Google has a product uh has a program called the associate product manager program which turns out people that look very much like product engineers.
21:51They turn out 50 of them per year.
21:54We need 10,000 of them per year.
21:57Even if you look at all of the other pro all of the other companies that have something a program like Google's APM, it's being extensively copied.
22:04They're only turning out about a thousand people a year and we need 10 times that number at minimum uh to keep up with demand.
22:11Uh
22:12[snorts]
22:12universities teach data structures, boot camps teach React, nobody teaches go sit with a baker for a week and come back with a spec and that is the person that we need.
22:21Um so if you lead an engineering team, you're going to have to grow them on purpose.
22:26Uh IBM is already redesigning its entry-level role around talking to customers and writing specs instead of programming.
22:32Um you're going to have to figure out how to do that for yourself.
22:35You're going to have to figure out how do you train product people?
22:38Uh and you're going to have to do it very quickly because the time we should have started was two years ago and there's a whole cohort of very sad unemployed juniors right now who have this problem and we are not hiring them and we need to do something about that or we are going to be screwed and
22:53not just them.
22:54Uh so if all of that happens, there's one big thing that I don't want to just gloss over or hand wave away, which is a lot of people got into programming
23:05because they love the craft.
23:07They love a clean abstraction, the feeling of a hard bug finally giving up, making a machine do exactly what you told it to do.
23:15They did not sign up to interview bakers about what they want in their lives.
23:19Like that is not the job that they signed up for.
23:22Um and they are going to hate this.
23:25They already hate this.
23:27They are complaining about it very loudly.
23:29I have no consolation prize for them.
23:32Um but uh I do have an observation, which is that for a lot of people who think they love the typing, maybe what you loved was the moment just before you started typing, when you figured out the problem.
23:44That That part doesn't go away.
23:46That part is still there.
23:48That creative spark when you're like, I know how to make this into a repeatable system that glows in my brain.
23:54That's not going away.
23:55Uh and the craft of programming will survive uh in the same way that like woodworking survived the furniture factory.
24:03Like programmers program the way that singers sing, right?
24:06They do it in an empty room with nobody watching just because they love it and that won't go away.
24:12You can still do that.
24:13Um it just won't be the thing that people pay you hundreds of thousands of dollars every year to do.
24:19Um so here is my prediction.
24:21The next 10 years are going to be ugly.
24:23The last 2 years have been ugly.
24:25Um the load of this stuff is arriving before the tools are ready.
24:29The junior ladder is gone before we have figured out how to train more juniors uh and a lot of people who love the craft are going to have to decide if they love the job that is replacing it.
24:39Um but this has happened before.
24:41Compilers and then frameworks and then open source each made a generations worth of typing before them uh unnecessary and each time we came out of it with more software developers.
24:52We just changed what software development meant.
24:55Uh and I think that's going to happen again.
24:58Uh, in 10 years time we will get a profession that is bigger than today's, pays about as well, and is mostly shaped like product engineering.
25:04That is the case that I'm making for you today.
25:08But that is a whole lot of high-level predictions, and I hate those.
25:11What does this mean for real?
25:12What should you actually do with this today?
25:14Uh, first, be prepared for a world in which humans are barely involved in the production of code.
25:19So, write down your high blast radius code, and make that be the only thing that you are reviewing anymore, because the rest of it your automatic reviewers are going to get done.
25:30Uh, then delegate that everything that's cheap to verify.
25:34Spend a lot of time thinking about how would I build an automatic way of telling whether this thing, whatever that thing is, is good or not, and then feed that to your agent instead of trying to figure out the review process.
25:45Uh, and finally, put your best people's judgment into specs and rubrics and evals.
25:50You know, I the arise guy was going to say evals at some point.
25:54Uh, get stuff that exists only in people's heads into code.
25:58That is your job as a product product engineer.
26:01It is turning the sense of the problem into a spec of the problem, so that the agents can build it for us.
26:08Uh, and measure the right thing.
26:10741% more code was never the point.
26:13Measure how much stuff you are actually shipping.
26:15Uh, measure how often your reviewers comments get acted on.
26:18Measure the gap between the tests [clears throat] passed and a maintainer would merge this.
26:22Uh, and remember that the last reviewer is production.
26:25Watching what it does there with another model grading it, that is an eval.
26:29I'm not going to pitch arise, that's just what we do, and that's why we're so popular right now.
26:34Uh, and I'll leave it at that.
26:36Um, 3 years ago, Swyx wrote about the rise of the AI engineer, which is engineers learning to build with models.
26:43And the rise of the product engineer is about what's left when the models do the building.
26:48Uh, finding out what people need, writing down what good means, and making it something people like, that is the job now.
26:54It is going to be the job for everyone in this room, whether we like it or not.
26:58We are all product engineers now.
27:01Thank you so much for your time and attention.
27:03[applause]