Why Humans Can't Keep Up With AI-Written Code - Tariq Shaukat, Sonar

AI Engineer · 16 min · 174 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]
00:13Uh I My name is Tariq Shakir.
00:14I am the CEO of Sona, as you just heard.
00:17And we're here to talk today about how we have been um helping companies build software factories that actually work for enterprises.
00:26And I thought it'd be an interesting way to start to start looking at actual factories cuz we know how to build actual factories.
00:33And for those of you who have a little bit of the gray hair and all of this sort of thing, you may uh remember the 1980s.
00:40And in the 1980s, you could find cars that were really good cars.
00:45But if you look at the picture here and if you remember those times, the cars like didn't quite fit together perfectly.
00:51Right?
00:51You'd have different pieces that would leave big gaps.
00:55It would be really good cars that just didn't kind of work the way they were supposed to.
00:59And then uh Toyota and particularly Lexus came out with a commercial.
01:04And it showed something very simple, which was a ball bearing running across the seams of this car.
01:10And it was kind of revolutionary at the time because it showed that you could actually build a car that fit together perfectly.
01:17And instead of building for volume and building and then fixing the old mindset, you could actually have a mindset focused on ship it right, on no rework, on customer happiness.
01:30And there was not This was not a single change.
01:34What happened was that people really started thinking about the factories differently.
01:40They started engineering a building factories that were engineered to deliver quality.
01:46And there's a number of different pieces that I think still resonate today.
01:50The first is you prioritize quality instead of volume.
01:54The second is that you fix the system, not just individual symptoms that you see.
02:01The third is that you design for something that can be built as opposed to designing the wheels and designing the washer bottles and whatever it is and then hoping it fits together.
02:12You solve problems as they occur as opposed to kicking the can down the road and fixing it at the end.
02:17And you empower everybody.
02:18Anyone could stop the line, not just the boss.
02:23Now, why am I talking about this?
02:24It's actually because modern software development actually embraced all of these principles.
02:28If you think about how we build software today, what is considered best practice, a lot of the terms and the practices came from that idea of a Toyota production system.
02:41Agile, Kanban, error budgets if you're in the SRE world, continuous integration, these ideas all kind of sprung from this notion of how do we build factories that work.
02:52And it's worked really well for us in software development so far.
02:55But agents, as we all know, you wouldn't be sitting here if you didn't understand that agents are changing the game.
03:02And as agents are changing the game, weird things start to happen.
03:07First of all, this is research that was published last month of the effect of implementing asynchronous agents in your development life cycle.
03:18And I'm sure many of you have seen this.
03:20You have an explosion of the number of lines of code written up 2,450%.
03:25You have an increase in the number of commits.
03:28So, the volume, code creation is now easy and it is now abundant.
03:34And by the way, more and more people, I'm sure many people in this room are saying, "Okay, if it's easy and abundant, how do I automate it?
03:40How do I build factories, agentic software factories that help automate and scale this?"
03:46And today, we just finished our what we call state of code developer survey.
03:51It'll be out next month, but this is a little preview.
03:54Today, 5% of developers that we spoke to say that they're running agentic software factories, but 34% said in a year they expect to and 39% in financial services.
04:05So, people are thinking about how do you actually build these next generation factories.
04:11And there's a a couple of gotchas in here.
04:13Security risk, the agents can work really well to help you build stuff.
04:17It can also work really well to help you hack stuff, help you attack stuff, find vulnerabilities, find quality issues, etc. And so, as I was writing this um presentation
04:28kind of over the weekend, a little just-in-time again from the Toyota production system, um Satya Nadella posted this article on X. If you haven't read it, I think it's a great article that I think everyone
04:39uh should internalize.
04:40But one of the things he was talking about insider risk of models inside of companies.
04:44And one of the things that he talked about was the need for an engineering approach to containment and governance built around strong deterministic system design, human controls, and reliable operating procedures.
04:58And this really [clears throat] resonated with me because what we are seeing is a paradigm shift in software factory design or software development, let's call it more broadly, that we need to engineer, we need to build systems that engineer confidence
05:14and quality into agentic software factories.
05:19So, to start doing that, I think it's important to understand the limitations that we're facing.
05:25So, what's the hard thing about this?
05:27And there's five that I'm going to call out.
05:28The first is that human in the loop doesn't scale.
05:32And this may be, I don't know, to this room it may not be all that controversial, but every virtually every company I talk to, I say, "What's your AI governance strategy?"
05:41And they give me all the stuff and they say, "But don't worry, we have humans in the loop."
05:44And the problem with humans in the loop, as I mentioned earlier, 2,450% increase in lines of code, 240% increase in commits, but the same study showed only a 30%
05:57increase in releases.
05:58We haven't figured out how to scale from producing stuff to shipping stuff.
06:04There's automation needed, there's new tools needed, and really importantly, this is a concept that really I I see it every day now, and I'm sure many of you do as well,
06:15uh called cognitive surrender.
06:17Uh and this study was done just a couple months ago, and it looked at how often do humans agree with AI?
06:24And it turns out they agree with AI when AI is right 93% of the time.
06:30And four out of five times they agree with it when the AI is wrong.
06:34All right?
06:34So, it's not that human in the loop is Yeah, we just need to make things faster.
06:39It's actually that it doesn't work in the agentic era.
06:45So, that's problem one.
06:46Problem two is that agents struggle with large complex code bases.
06:51They're very very smart, but you need to understand what happens in a code base in order to work on it.
06:58The left-hand side is a study that was done on code translation, and what it shows is that
07:03[clears throat]
07:03the agents were able to get it right 91% of the time when the code base was 10,000 lines of code.
07:12And by the time it got to even 50,000, and I'm sure everyone in this room operates off of much bigger code bases than this, but when it was trying to translate 50,000,
07:20the accuracy collapsed.
07:23Similarly, another study looked at um how many tokens does it take on a large complex code base to complete the same task, and the variance was 30x, not 30%,
07:34but 30x, and the reason is because it the agent spends a huge amount of their tokens trying to understand how to navigate the code base, the structure of the code base, things like this.
07:45And it's kind of like if I'm trying to go from uh from LA to here, and if I pick a random starting point.
07:53If I happen to pick White Plains, I will get to New York pretty quickly.
07:57If I pick Albuquerque, it's going to take me a long time and a lot of cost.
08:00Same thing happens with agents.
08:04Errors compound.
08:05This is another important concept.
08:08This is a study out of meter and meter looks at and you can find this on their website.
08:14They look at how effective are agents at completing tasks that humans can do.
08:21And the way they look at this is can you complete a task that takes X number of hours.
08:25So you can see here the Claude Mythos preview which is the latest version that they've published was good at completing tasks that take a human 16 hours.
08:34Okay, so that's really good but there's an important caveat in the fine print here which is this is at 50% accuracy.
08:41When you add it when you dial up the accuracy to 80% it the curve flattens substantially because as you do long running tasks errors compound.
08:52If you have 95% accuracy on each task by the time you're at 20 steps, you have 30 36% accuracy.
09:01Right?
09:01And so that is another issue we have to deal with in software factories.
09:06Probably the most aggressive headline I put in here is is this fourth point that LLMs are unreliable reward seeking judges.
09:14Um as you know, so the question that we always get is well can't you just ask an LLM?
09:19And so people have been studying this as well and this is all again the first on the left hand side is from PayPal AI's research lab.
09:27They published a paper this month actually and found that between depending on the model and the situation between 7 and 17% of verdicts change across what they called identical replications.
09:40Ask the same question, got a different answer between 7 and 17% of the time.
09:45And if you start challenging, this is a study out of meta now.
09:49If you start challenging the LLM, are you sure?
09:52Very simple challenge.
09:53It'll actually change its mind 7 to 34% again depending on the model.
09:58So you ask the AI judge twice and you may get a different answer.
10:03And then lastly and I won't dig into this one too much cuz I think it's probably self-evident to everyone as the time to exploit near zero, the ability for us to respond
10:14is incredibly challenged.
10:16And so, prevention is something that we believe the software factory has to bake in to the design.
10:22Has to engineer into the factory.
10:25So, as we've thought through this, we have um written down what we call the code confidence formula, which is basically how do you build How do you think about
10:38a software factory that delivers what we all want, which is outcomes that we can feel confident in, right?
10:45Not outcomes we can trust, not blind trust or anything like that, but outcomes that we can feel comfortable with or we can sit and tell our executives, our customers, our shareholders,
10:54our CEOs, our boards that yes, I believe this factory is working well.
10:59And we think there's four components to this.
11:02The first is being crystal clear about the intent.
11:07If you don't know what you're trying to build, you can't tell the agent what to build, and you will not get the outcome you want.
11:12The second is guide rails.
11:16And we very specifically talk about guide rails, not guard rails.
11:19Guard rails prevent something from going off the road.
11:23Guide rails help keep you on it.
11:24And I think what agents need are the small nudges to guide and govern those agents.
11:30The third part is verification that needs to be independent, thorough, and issues fixed as opposed to just cans kicked down the road.
11:39And then lastly, evidence.
11:42Trust, but verify.
11:43Trust, but understand that it was done with proof.
11:47Uh that is consistent, audible, and repeatable.
11:49And we multiply these together because we've seen a number of situations where one of these elements is missing.
11:56And if it's zero, the confidence goes to zero.
11:58If you do all of this, but don't have evidence, you go to zero.
12:02You don't have confidence in the outcome.
12:05So, we really think about factory design is requiring these four elements.
12:11And I'm going to And as we put together factors, as we help customers, I'm not going to walk through every box on this page, but as we help our customers
12:19and partners build these software factories, we're very deliberate about how do you get the intent, bottom left-hand side in purple.
12:27How do you put the guardrails in place?
12:29How do you do verification in a way that's thorough and independent?
12:33And how do you produce the evidence?
12:35And by doing this, we see amazing results.
12:39So, these are real results now from our from
12:41[snorts]
12:41our customers.
12:42The first one is focusing on intent and guardrails.
12:46And intent and guardrails, we have a product called Sonar Vortex, which gives intent guardrails and in the agentic loop starts to do verifications, again nudging the agents.
12:55And what we see when we provide agents in long-running and large complex code bases with long-running tasks uh with
13:04[snorts]
13:04this information is a 36% reduction in token cost because you're putting it in White Plains, you're actually you're putting it close to the answer, you're helping it understand the map.
13:15You don't have to burn tokens on things we already know.
13:18And 92% reduction in issues in the commit.
13:22Right?
13:22So, it's cheaper and we can get a better outcome to go into that verification stage.
13:28This is again all because we're nudging the agent to keep on track.
13:33Then you get to verification, and we started life as a verification company.
13:38This is how we see the world as through a verification lens, and we have a very strong point of view that independent verification has to cover the full range of issues
13:48that is both precise and cost-effective.
13:51Right?
13:52Use CPUs when you can use CPUs, need GPUs GPUs when you need GPUs.
13:56So, we look at deterministic mixture methods, and then you add in agentic mixture methods, and it comes up with what we think is the a gold standard of verification.
14:07And what we see is when you apply this, our product is called SonarQube.
14:12So, we talked to our customers, over 1,000 customers recently, and we said, "Are you How likely are you or how often have you seen these issues from a gentle coding?"
14:25And the the statement to read here is SonarCube users are X% more likely than non-users to have seen a very positive impact.
14:32So, 81% more likely to have seen to believe that AI coding has had a very positive impact on code security if you have that gold standard ver- verification.
14:4534% more likely to have seen an impact on time to market, 76% on outage frequency.
14:54So, again, as I come back to how do we have confidence to ship, what we're seeing is that if you put these pieces together and you build verification incredibly
15:04strongly as a first-class citizen in the factory, you are actually building confidence and you're seeing results, better security, more confident in time to market, improved or reduction in number of outages, improved reliability.
15:21And, of course, as I mentioned, you have to do this with evidence.
15:25Make sure you pass the quality gate.
15:26Make sure whatever There's lots of acronyms up here, whichever ones you need to pass, whether that is your IT general controls inside of your internal controls and operations, whether that's MISRA if you're building cars,
15:39or any of the other ones that are on here, producing the evidence is what lets you sleep at night.
15:45It's what gives you confidence that you can move forward.
15:49So, that, in essence, is how we think about the challenge of the future, that code is abundant, that confidence is scarce, and really the goal that we have and that we think
16:03that everyone should have as we think about the future of software development is how do we deliver confidence.
16:09And the factories of the future need to be able to deliver confidence.
16:14So, with that, I want to say thank you to everybody.
16:16We have a big booth, it's a big red booth on the floor right over that way and would love to talk to you more about all of this.
16:25Thank you very much, have a great morning.
16:27[applause]