“That’s the intelligence explosion. Once AI can iterate and self-improve then it is just a function of how many agents we can throw at that self-improvement and then it is exponential by nature.” — Matthew Berman
our bet is sort of that in the next year probably you know I don’t know maybe half the development is going to be done by AI as as opposed to people and then that will just kind of increase from there mark Zuckerberg and Satia Nadella the CEOs of Meta and Microsoft respectively got together for a fireside chat at Llamicon 2025 and they covered everything from open-source to agents to how coding is going to look very different in the next year and a half let’s watch it i’ll give you my thoughts all right so we’re going to go a little bit out of order i want to go straight to the part where they talk about AI coding and specifically how much code is written for their respective companies by AI already and what the future might look like because it is fascinating in terms of the coding and and how it improves that do you have a sense of how much of the code like what percent of the code that’s being written inside Microsoft at this point is written by AI as opposed to by by the engineers yeah so there’s two sort of things we’re tracking one is the accept rates itself right that is sort of whatever 30 40 it’s growing up monotonically uh and it depends like one of the big challenges we had for a long time is we are a lot of our code is still C++ um C and C# is pretty good but C++ were it was not that great Python it’s fantastic so we now gotten get better at that so as language support has increased the code completions have gotten good that’s because the models aren’t trained on C++ as much obviously there’s some training data there for C++ ++ but I don’t believe C++ is open- source and certainly if you’re writing C++ you’re using a lot of closed source tools but Python on the other end is open source the tools you use are open source the libraries are open source and that has made these models really good at being able to write Python because simply there’s a lot of it to train on the place where the agentic code still it’s very it’s sort of nent for new green field it’s very very high uh but as I said it’s nothing is green field uh in many cases and so therefore I would say maybe at this point the PR oh by the way code reviews are very high so in fact the agents we have for reviewing code uh that that usage is increased and so I would say yeah I want to pause here for a second because it’s very interesting he says basically nothing is green field or very few things are green field but that’s only true if you have a massive legacy company with a massive amount of legacy code microsoft is run on code that was written decades ago so not only are the agents not as good at writing those languages but there’s just so much code so much dependency between the code that it’s hard for these agents to understand and actually iterate on those types of code bases the context is just too large at least for now but that doesn’t mean that there’s another company a startup building something from scratch building code in a way that is especially written to be iterated on by AI that isn’t going to be able to accelerate to a pace much faster than Microsoft can even imagine so although there are very few green field opportunities for Microsoft that doesn’t mean there aren’t many green field opportunities for everybody else he also says that code reviews are a big part of what they use AI for and that would make sense because with code reviews you’re taking a subset of the total amount of code you’re seeing the differences the diff between the new code and the old code and you’re asked to review it does it make sense does it make sense within the context of the related code around the code you just changed and so that does seem like something AI could do incredibly well but now he’s going to talk about specific percentages and keep in mind usually there’s no distinguishing between code completion which is done by AI so you start typing a line of code a function and you just hit tab and it completes that’s technically AI written but if you are completely relying on AI to write code for example you’re assigning a PR to an agent I would consider that kind of a different level of AI coding but I think it’s all mixed together in this number he’s about to say that usage is increased and so I’d say maybe 20 30% of the code that is inside of our repos today in some of our projects are probably all uh written by software what about you guys um I actually don’t have the number off the top of my head but I mean it’s um I I I think you know we I think a lot of the stats that people say are still effectively of this like autocomplete variety yeah but we have a bunch of teams that are working on um on basically doing feed ranking experiments and ads ranking and like very contained domains where you can study the history of all the changes that have been made and like and and and make a change and that I think is like is kind of an interesting area for for us to work in but the big one that we’re focused on is um building an AI and a machine learning engineer to advance the llama development itself right because and what he just said is probably subtly the most important thing in this entire talk they are developing artificial intelligence that can do ML and AI research again and I’ve said it a thousand times that’s the intelligence explosion once AI can iterate and self-improve then it is just a function of how many agents we can throw at that self-improvement and then it is exponential by nature and let me just pause this conversation to tell you about today’s sponsor Outskill outskill is a live 2-day AI training program for professionals founders and executives through this live 2-day program you will master the skills of AI including the basics of generative AI automations building AI agents image and video generations generating full-fledged websites and more the 2-day training happens Saturdays and Sundays from 11:00 a.m to 7:00 p.m eastern and there’s an initial kickoff Friday 10:00 a.m 2 days 16 hours 5 sessions you will learn so much 50,000 professionals have already attended these sessions over the last 6 months and they’ve landed consulting gigs built AI products or just upskilled themselves in their existing roles they also offer live Q&A sessions with mentors so you can ask the questions and clear up any doubt that you might have and clarify any topics that might be confusing so check out Outskll i’ll drop a link down below it’s free for the first 10,000 people who register and thanks again to Outskll now back to the video all right so let’s keep watching and think about the possible intelligence explosion that he is really describing here but the big one that we’re focused on is um building an AI and a machine learning engineer to advance the llama development itself right because I mean our our bet is sort of that in the next year probably you know I don’t know maybe half the development is going to be done by AI as as opposed to people and then that will just kind of increase from there so I I was just curious if you were if you were seeing something different yeah I mean to me that the the uh agent is the sort of the first attempt so the question for us is in the next year can we get um like let’s take a kernel optimization right will we get to sort of something like that that happens I think it’s more likely whether by the way when he’s saying sweet agent he’s saying software engineering agent just think agent that writes code for you comes up with a novel model architecture change probably not so the question is which task yeah no optimizations security improvements that type of stuff I think seems like it’s it’s pretty high opportunity Honestly it sounds like Zuckerberg might be thinking more in the future than Satia is unless Satia is playing his cards close to the vest now Microsoft isn’t really developing their own models they’re partnering with everybody else meta is developing their own models he’s probably thinking about making these models able to self-improve maybe more so than a Microsoft would be yeah yeah no I we’re also trying to solve a different problem on it because I mean you guys serve like a lot of developers and engineers that’s like your core business um whereas for us we’re thinking about this more as a thing to improve our internal development and then improve the llama models which other people can use but it’s not something that we do the endto-end workflow on in the way that you do so it’s always just interesting to hear how you’re thinking about that yeah I don’t know maybe that’s going to be changing i think they might be both trying to deceive each other or at least they’re definitely not telling the full story because Meta just launched an API so they are kind of creeping in the direction of serving developers providing inference and uh yeah that’s Microsoft’s territory the other thing for us is yeah to your point our core business in fact you know Bill started the company as a tools company and so to us uh the interesting thing I’ll think about now is maybe the way we we should reconceptualize our tools is and infrastructure quite frankly are the tools and the infrastructure for the agents to use because even the three agent needs a bunch of tools that’s the key the entire technology stack needs to be rethought for the world of agents how will agent agents write code how will they retrieve things from a database how will they browse the web how will they talk to each other how will they do tool usage all of these things need to be standardized they need to be figured out it’s so early and very exciting for that reason but he’s exactly right we need to figure out what the world of software and beyond looks like when we’re thinking about AI first and what shape should they be uh what should their infrastructure what should their sandboxes be so a lot of what we’re going to do uh is essentially evolve even what does the GitHub repo construct even look for the su agent yeah no it’s that that’s a it’s a very interesting concept and I mean I I tend to think that like like every engineer is effectively going to end up being more of like a tech lead in the future that has sort of their own little army of of of engineering agents that they work with but yeah that is the orchestrator that I was describing a tech lead orchestrating tens hundreds of different agents how these agents work together to execute on the task assigned to them by the orchestrator is yet to be determined and we do need to rethink what a repository looks like because instead of a human team of five 10 20 people working on a single repo on different branches now you could have 5 10 hundreds of agents working on a single branch at the same time in coordination i don’t know i really don’t even know what it’s going to look like maybe that’s completely off let me know what you think by the way tell me in the comments next they’re going to talk about what new developers if you are getting started in the development industry what should you be thinking about how should you be thinking about AI coding and everything else involved with writing software and I get this question personally all the time people asking me should I learn to code is it still worth learning to code i think for now yeah but in the long run knowing how to do the actual base code might not be as relevant i think it’ll be cool to know but I think it’ll be something artisal almost but agents are going to be writing the vast majority of our code so it’s more about learning how to interact with your AI agent team and also I think that it is incredibly important to learn how to do systems thinking so learning how to code teaches you systems thinking you don’t necessarily need to know any individual programming language but learning that systems thinking first of all even if you never write a line of code is just valuable in life but even if you’re orchestrating agents systems thinking is important i guess there’s always this question if you were getting started as a developer today building something how would you think about which tools you’d be using and and um yeah I think that one of the one of the biggest sort of I’ll call it I don’t know dreams pursuits questions that Bill sort of inculcated in all of us was what’s the difference between um uh he used to sort of talk about it more like what’s the difference between a document and an application and a website right now if you use meta chatgpt copilot what have you it’s unclear to me what’s the difference between a chat session Mhm uh and then then I go to pages in our case like literally even coming down you know I was like reading up everything about Llama 4 all the models like literally I was just doing a bunch of chat sessions adding it to a document effectively in pages persisting it uh and then you can go give it I mean since you have code completion you can go you know make it an app or what have you and so this idea that you can start with a highlevel intent and end up with what is an artifact that is a living artifact that you would have called in the past an application is going to have profound implications I think on workflows and I think we’re at the beginning of that uh and that’s what my dream is if I sort of say as a builder of infrastructure and tools and as a user of it these artificial category boundaries not artificial or these category boundaries that were created uh mostly because of limitations of how our software worked perhaps you transcend um in fact the other thing we used to always think about is why are word excel powerpoint different why isn’t it one thing and we’ve tried multiple attempts of it uh but now you can conceive of it right which is you can start in word and you can sort of visualize things like Excel and present it and they can all be persisted as one data structure or what have you and so to me that malleability that was not as robust before is now there yeah interesting makes sense yeah so I think what he’s getting to is the interface between humans and data and he’s saying there’s the malleability between these different pieces of siloed software but really what I think it’s going to end up being is there’s going to be an AI agent between human and the ground truth data i think that is the perfect architecture for the future the ground truth data is going to be stored in a traditional database maybe it doesn’t look like a database we know today but it’s going to be a persistent database it is going to be deterministic by nature that is the ground truth then there’s going to be an agent and then a human on the other side the human is going to need to do something learn something figure something out assign something whatever it is and they’re going to talk to the agent the agent is going to do everything else it’s going to write all the software it’s going to retrieve all the data necessary to accomplish the task assigned to it with that in mind there really is no software and I’ve said this before i know I’ve said software is dead a little bit tongue andcheek but if you really think about it there’s going to be a database there’s going to be an agentic layer and then there’s going to be the human on top orchestrating everything and at that point where are the applications i don’t know i don’t know if there are any let’s keep watching all right next Mark asks Satia about how he thinks about the actual application of artificial intelligence he prefaces with Satia saying “Well all of this infrastructure investment all of this innovation in AI needs to be reflected in GDP growth.” And of course it does as a capitalist society the investment actually needs to have a return or it’s going to be a failure and so he asked him “How do you think about that?” So let’s watch i’m curious how you think what’s your current outlook on on sort of what we should be looking for to understand the progress that this is making and and how we would and and kind of like where you expect that to be over like a three five seven year period to me I think that that’s right because to us I would say it’s a pretty existential priority quite frankly the world needs sort of a new factor of production and input that allows us to deal with a lot of the challenges uh we have um and uh and the best way to think about it is hey what would it take let’s say for the developed world to grow at 10% um right which may have been some of the peak numbers during uh let’s say the industrial revolution or what have you um and for that to happen then you have to sort of have productivity gains in every function right in health care in uh in retail uh in broad knowledge work in any industry and for that to happen that I think AI has the promise but you now have to sort of really have it deliver the real change in productivity and that requires software and also management change right because in some sense people have to work with it differently uh you know people always quote what happened with electricity right it was there for 50 years before people figured out that hey we got to really change the factories to really use electricity differently right and that was the the famous Ford case study and so to me we’re in somewhere in between I hope we don’t take 50 years um but I do feel that by just thinking of this as whatever the horseless carriage uh is also not going to be uh the way we’re going to get to the other side so It’s not just tech tech has got a progress uh you got to put that into systems that actually deliver the new work work artifact and workflow so I don’t think I can say it any better than he just said it there are problems and AI has the promise to solve a lot of these problems that are existential to not only the United States but to the earth as a whole so you know you’ve said a number of times that this moment in technology around um the the growth of AI sort of reminds you of of some of the important transformations in the past from going to client server and the beginning of the web and things like that so I’m I’m curious to Yeah so for me u it’s interesting right which is each time there is this transition uh everything of the stack gets relitigated um and you get to sort of u go back to the first principles and start building uh I mean I I I thought like even the the shape of the cloud infrastructure uh for me that I built starting let’s say in 2007 8 to what have you the core storage system for training doesn’t look like the core storage system you built or this this workload of training right the data parallel synchronous workload is so different let’s say Hadoop or what have you so the the fact that you kind of have to rethink everything um up and down a tech stack uh with each of these platform shifts is sort of what I think we face from time to time it sort of grows from what was there uh the web was born on Windows but it went far beyond that that’s kind of how I think about this as well so to me this is the most exciting thing about this new wave that we’re going through right now and personally I believe it’s the biggest wave of innovation and technology in history but basically we get to reimagine what the infrastructure of everything looks like because of artificial intelligence we’re in this kind of awkward phase right now and typically when a brand new technology comes out everybody scrambles to try to fit it into the way that things already worked the existing patterns that everybody’s used to so a couple of examples of that in the early days of the internet everybody just rushed to take what was an existing pattern print media and just literally slap it onto a website and what they didn’t realize was there was this entire interactive layer that was possible and so many other elements that you just couldn’t have with print media that was possible by having the internet and I think AI coding is a great example right now we’re using AI to help us code but using our traditional patterns a lot of the tools we’re using are essentially traditional IDE with AI kind of slapped onto it and now we’re starting to get more we’re getting tools like cursor replet winds surf and these tools are starting to head in the right direction where we’re thinking about things from the ground up what does coding look like in the future where it is almost entirely done by AI and the human is the orchestrator and so again we are just scratching the surface of what is possible and the exciting thing is we get to invent this new pattern for how AI interacts with basically everything in our lives all right let’s keep watching yeah that makes sense i mean you’ve you’ve made this point a bunch of times around how as things get more efficient um it it sort of it changes the way it works and people just end up consuming a lot more of the services right yeah so he’s talking about Javon’s paradox as a resource becomes cheaper the intuition is that people will spend less on it because if demand stays the same and prices are less then the total price paid will also be less but with Javon’s paradox what it shows and this was easily shown with DeepSeek R1 the cheaper something gets the more efficient something gets it actually has the opposite effect although the per unit cost goes down the total consumption of that resource goes up all right let’s keep watching how how you’re seeing that play out around um around all these AI models right you’re seeing like generation over generation they’re just getting so much more efficient and delivering more intelligence than the last generation and you know obviously it’s all happening super quickly so I’m not sure kind of how what what you can what you see in that but yeah I mean if you think about it right which is you know we were all you know a few years ago sitting around and saying oh my what’s happened to Mo’s law you know is it over what do we do and here we are in some crazy sort of hyperdrive Moors law and it’s always been also the case right which is any one of these tech platform shifts has not been about one scurve it’s been multiple scurves that compound right even if to take uh just the fact that the chips are getting better uh you know people like Jensen or Lisa doing tremendous innovation jensen and Lisa are cousins don’t forget that jensen CEO of Nvidia lisa Sue CEO of AMD you know their cycle times have gotten faster so let’s say that’s Mo’s law but on top of that everything at the fleet the system software uh optimization uh the model architecture optimizations uh the kernel optimizations for inference the app server even the prompt caching how good we’ve gotten in uh and so you add all of that up for every 6 months 12 months you have a 10x perhaps improvement right and so when you have capability improvement of that rate and the prices drop at that rate fundamentally consumption goes up so I’m very optimistic uh that we are at a stage where deep applications can get built yeah and keep in mind when he says applications he is talking about the software built on top of the infrastructure and model and hardware layers and I think that is probably the biggest opportunity today is building applications on top of these models and I still think there’s a lot of opportunity to build the infrastructure the scaffolding around the models i just posted about that on Twitter by the way if you don’t follow me Matthew Burman on Twitter things like memory management coding assistance agentic frameworks tool use prompt management and optimization there’s so much to do there but then on top of that the thing that really we’ve seen very little of to date is the application layer how are we going to deliver all of this amazing AI to consumers and to enterprise that is still being figured out that is probably the least explored area of the AI stack to date where you have an orchestrating orchestration layer with these agents with multiple models i feel like we’re at that place because if you think about even the first generation of apps they were very coupled uh very coupled to one model but we are finally getting to multimodel applications where I can orchestrate in fact a deterministic workflow an app agent that was built on one model talking to another agent we even have these protocols are helpful whether it’s MCP whether it’s A2N whatever these are all good things if we can standardize a bit and then uh we can build applications that are you know taking advantage of I would say uh the you know these capability building but has flexibility and that’s where I think open source absolutely has a massive massive role to play yeah so he just mentioned a lot of things that I’m very interested in agentic frameworks where you can coordinate multiple agents together being able to have each of those agents powered by individual large language models even from different providers let’s say you have one from OpenAI you have another from Meta you have another from Anthropic and they’re all talking to each other you’re finding the best model for the specific task at hand and then even within that even within each individual prompt you have things like model routing and you can actually optimize on a per prompt basis for the cheapest most efficient and best quality for each individual prompt very excited about all of these things but still the application layer I think is the biggest opportunity right now yeah well well I definitely want to make sure we can get into discussing how to use multiple models together and I think that there’s this whole kind of concept of a like distillation factory and the information and the infrastructure around that that that you think Microsoft is well positioned to basically provide is there are multiple models maybe we’ll come back to that in a minute but before we do that you know Microsoft obviously has been on this interesting journey around open source right and this is one of the big things that you did you know under your leadership early on was embracing it and you know you had the early partnership with OpenAI but then also were very clear that in addition to working with closed models you wanted to make sure that Microsoft served open models well yeah that is something that I think was the 4D chess move by Satia he identified very early that open AI was an extreme dependency for Microsoft if they relied on open AI for their models and only open AI i think he noticed either with Sam Alman or OpenAI more generally they had bigger ambitions than just being a model provider and specifically a model provider to Microsoft and you could start to see that with the Windsurf acquisition rumors you can see that with their push to building consumer applications and so Satia diversified away from that i’ve talked a lot about platform risk on this channel this is exactly that and I just think his investment into multiple different closed source and open source model providers will be looked upon as one of the greatest strategic corporate moves in history and I’m I’m curious how you think about that and how you think that the open- source ecosystem is going to evolve and why that’s important to um your customers and how you think about that with all the infrastructure that you’re building yeah it’s it’s it’s interesting you ask that because I I I grew up um in fact one of my formative jobs at Microsoft was also making sure uh that we had interoperability with the various flavors of Unix out there between NT and Unix uh and that taught me one thing which is interoperability is what first of all customers demand and if you do a good job of it um that’s good for your business and obviously you’re meeting customers where they are i’m not dogmatic about closed source or open source both of them are needed in the world uh and in fact I think customers will demand them right even if any one of us has dogma it doesn’t matter because at the end of the day the world will break that way which is there will be a need for it it also fits with what you just talked about because a a lot of my enterprise customers want to be able to distill in many cases models that they own it’s their IP uh so that in that place where an openw rate model has a huge structural advantage i’m very surprised to hear that i’ve had a lot of conversations with business leaders who are thinking about the ways that their company is going to adopt AI and it has been very seldom that they are saying “I want to fine-tune my own model i want to distill my own model.” And I’ve said it on this channel before nine times out of 10 fine-tuning is not necessary for your business usually rag retrieval augmented generation is the answer making sure that you’re giving the right context to the model at the right time all right next they’re going to talk about a topic very near and dear to my heart agents so let’s watch yeah so so how um you mentioned agents and and increasing productivity and that’s obviously a huge theme for for the whole ecosystem and community i’m I’m curious how are you seeing that play out inside Microsoft and then also how are you seeing that um what are some of the most interesting examples that you’re seeing with development yeah I mean I think the the the thing that obviously has been most helpful for us to see is what’s happened with software development right i mean there are a couple of things if you look at even the evolution of GitHub copilot you started with code completions then you said let’s add chat so that that means your you don’t need to go to Reddit or Stack Overflow and you could stay in the flow so that was good then uh the agentic workflow so you could just go assign a task um so those three things if I look at even any one of us using it you’re using all three at all the time right so it’s not like one substitutes the other and now you have a proto agent even uh and so you can literally go highle prompt or you can just get a PR assigned to a sui agent so all four of those and the productivity gains the biggest lesson learned there Mark is you got to integrate all of that with your current repo and your current developer workflow right I mean it’s one thing to build a new green field app but none of us get to work on complete green field all the time right so obviously that’s a very good point there’s a lot of existing code a lot of existing processes and teams out there so we need to build all of this AI infrastructure for coding directly into the processes that people are used to But personally I’m more excited about that green field i want to think about what does the future of coding look like when it is almost entirely if not entirely done by agents what does the IDE look like what does that orchestration interface look like it likely doesn’t look like VS Code to be honest vs Code looks the way it does because we as humans need to be able to read and write code in the easiest most efficient way possible but when agents are doing that maybe the interface can be completely different what does that interface look like for an orchestrator and even beyond that what does code look like when it doesn’t necessarily have to be written by or for humans if agents are writing the code agents are reading the code the computer is executing the code why does it need to look like natural language at all and it kind of does even though we have all of this syntax around it ultimately when we look at it it kind of resembles natural language now it’s that way because humans are really bad at writing symbolic code for example so I’m really excited to think through what that looks like okay let’s keep watching there’s no need for somebody to prepare anything just because it’s all available on tap so it requires you to change work work artifact and work flow and that’s where that’s a lot of change um and it happens slowly at first and then all of a sudden here’s what cursor’s Aman Sanger had to say recently cursor writes almost 1 billion lines of accepted code a day that is absolutely insane to think about and that is just cursor that’s not including Replet Windsurf Klein and the other AI coding tools out there to put it in perspective the entire world produces just a few billion lines a day so cursor already is a substantial percentage of all of the code written and this is exactly what I’ve been talking about for a while we’re going to see this absolute explosion in the amount of code written every single day and more importantly we’re going to see an explosion in the amount of people who can write code because of things like vibe coding because of being able to describe what you want to see built from absolutely nothing in natural language to something that actually manifests and you can deploy and other people can use it you can use it and that is so exciting to me to go from I don’t know actually how many developers there are in the world maybe a few million to literal billions of people who can write anything that they want all right so that’s all I want to cover for today there’s a little bit towards the end about model distillation and a call to developers to build meaningful functional tools for the real world