(Illustration: Layer upon layer, every piece matters. Like this ham air-drying in the breeze, sacrificing itself for the greater good. Trust can’t be built all at once; it has to be stacked, one layer at a time. The first layer? Start by hitting that like button. Image source: Ernest.)
✳️ The Boring Plumbing Work
What actually blocks enterprise AI deployment is almost never that the models aren't good enough. Mistral AI CTO Timothée Lacroix put it bluntly: today's model capabilities are already sufficient to unlock massive enterprise value, but you first need to get all the connectors, data formats, and permission management right — all this "boring plumbing work" — before enterprise token consumption truly takes off. He used the word “plumbing” (and I’d stress the “system” part, not just the pipes). He said it three times throughout the interview. We’re still in the construction phase, he noted — most enterprises haven’t even gotten basic data connectivity right, let alone running AI agents executing tasks at scale in the background. (What we see on the ground is even bleaker: wishlists that don’t map to actual data. No wonder everyone just jumps straight to word-chaining chatbots — sarcasm mode engaged.)
So where should enterprises start? Build trust first, then talk about autonomy. When Lacroix was asked about agent autonomy, he flipped the question to something more fundamental: “Rather than asking how autonomous AI agents should be, a better question is how much you trust them.” Their approach working with a shipping company to automate container release processes was concrete: AI automatically handles all data checks and back-end verification, collecting information scattered across different systems, but since each container is extremely high-value with zero tolerance for error, final decision-making stays with port personnel. This is the first rung of the trust ladder: let people see what the AI did and why it did it, confirm that the AI’s judgment is traceable and verifiable, and only then does trust have a foundation to stack upon. Once trust accumulates to a certain level, you can gradually let agents execute more tasks in the background.
At PAFERS and Kyklosify, we follow a similar logic when working with customers using our hands-on approach and working backwards methodology (a.k.a. consulting the oracle). We start from the customer’s concrete pain points and reverse-engineer: what product is needed, what processes, what systems, how to align stakeholders, how to reduce the cost of understanding and digesting, and how to fine-tune organizational culture. You can’t pick tools first and then look for problems. Instead, you identify the root and nature of the problem, play board-game-style simulations of data flows on a big table or whiteboard, and co-design solutions together. We choose to work directly with organizational decision-makers whenever possible, because what stalls most often during iteration cycles is decision-making, not technology (it’s the guts, not the brains). Trust can’t be built all at once; it needs to be stacked layer by layer, one on top of another.
Engineering context is the next critical bottleneck. Lacroix mentioned what they internally call the “context engine” concept: knowledge that agents discover while exploring enterprise data — such as which data tables exist, how columns join, what permissions grant access — should all be stored and reused, rather than having agents start from scratch every single time. He said: “These computational costs should be amortized.” Imagine this: an agent spends five API calls and three joins to locate a piece of data. That path should be remembered and reused directly next time. Right now, every time means walking the same exploration path all over again. That’s unreasonable. (What’s even scarier is when different paths don’t even converge to the same result.)
This echoes what I keep reminding myself: “Coding is Easy, Context is Hard.” Once the engineering bottleneck is unlocked, the bottlenecks of product thinking, human communication, and contextual understanding are just getting started. Code output velocity has skyrocketed, but the real bottleneck isn’t just writing code — it also includes managing and passing context. Whoever can systematize, operationalize, and parameterize an organization’s tacit knowledge holds the moat of the AI era (for now?) (or maybe it’s not a moat, just a temporary bluff — don’t forget to deploy the unassuming sweeping monk). When we were quietly building the Kyklosify Business Suite over the past few years, as diehard AWS fans, we modeled our approach after Bezos’s 2002 API Mandate: working alongside customers to decompose every workflow in their organization into API-callable actions, ensuring every operation has the opportunity to leave a historical record. This groundwork paves the way for future Agents and Digital Twins. This upfront investment ensures that process knowledge can be systematically preserved and reused by other agents or humans, without starting from scratch every time. This foundational groundwork has enabled our clients to serve 80%+ returning customers, invest in building innovative services, and achieve revenue growth of +200% year-over-year (3x compared to last year).
Building your own infrastructure is a pragmatic choice, not a luxury (but please don’t attempt it lightly without the resources to back it up) (pragmatism comes in many flavors). Mistral’s decision to build their own data centers wasn’t a political statement about European sovereignty — it was because running training tasks across thousands of GPUs on someone else’s environment didn’t meet their stability requirements. Their core value proposition is control: the software stack, once deployed, belongs to the customer; the right to modify models also belongs to the customer. The entire stack is modular — customers can choose to use just the model, just the platform, or full managed service; every layer is theirs to decide. Lacroix even said that even if AGI arrives, banks won’t let it control everything — infrastructure governance capability must keep pace with model advancement. Having used AWS for over 18 years, I understand the logic behind this choice: the point isn’t a binary between self-hosted or cloud, but ensuring that at every juncture, “when I need to change, I have the option to.”
If you’re driving AI adoption in your organization, Lacroix’s experience points to a clear starting point: inventory the processes in your organization that are highly repetitive and rule-based, and start building the trust ladder there. Don’t start with the coolest use case — start with the most boring one that can be most easily verified.
The boring stuff is where value hides best.
✳️ Further Reading
- Coding Is Just Typing, Then What? From Explicit Programming to Implicit Programming
- From Vibe Coding to Agentic Coding: Clear Communication Is the New Bottleneck
- 2025 Year in Review: Slow Down to Go Fast
- AWS Summit Hong Kong 2025 Dev Lounge: Reinventing Programming - How AI Transforms Our Enterprise Coding Approach
- Amazon Bedrock AgentCore Goes GA: Enterprise-Grade Infrastructure for Production AI Agents
✳️ Knowledge Graph
(More about Knowledge Graph…)
✳️ Transcripts
Enterprise Demand and Control
- I think the expectation is that demand and amount of tokens generated for the enterprise will completely jump once you are not bound anymore by humans asking questions or reading them.
- As soon as you have enough trust to have agents running in the background, you’re not really limited by the number of tokens.
- The term we use is control.
- The software stack once deployed is in the hands of our customers.
- They own the model changes that we make.
- And I think it's really important as a customer to consider that your expertise and what makes your company valuable stays yours.
Podcast Introduction
- » Hi, I’m Matt Turk.
- Welcome back to the Mad Podcast.
- Today we have a special episode with Timote Lacro, the CTO and co-founder of Mistrol, the company that proved that you could build frontier models with a fraction of the compute of the US giants.
- But recently, Mistrol has quietly evolved into a much more ambitious full stack industrial power, building not just the models, but the platform, the deployment stack, and their own massive supercomputing clusters.
- We covered a lot of ground in this one, the engineering behind Mistral 3, what sovereign AI actually means in practice, and Tim’s contrarian view on why trust matters more than autonomy for agents.
- If you’re tired of the AI hype, Tim is refreshingly nononsense.
- Please enjoy this great conversation with Timote LRA.
- » Hey Timote, welcome.
- » Hey.
Shift to Full Stack Solution
- » So, as I was prepping for this, I was struck by how much has been going on at Mistrol over the last few months.
- I think most people probably know Mistrol as a provider of open-source models.
- It seems that you guys evolved from an AI lab to more of a full stack solution focused on enterprise and sovereign customers.
- So just to set it up, in the last year you guys raised a€ 1.7 billion euros series C led by ASML at an 11.7 billion post money valuation.
- you launch a bunch of models which we’re going to talk about is the big vision behind all of this that enterprises and sovereign states are going to need their own AI infrastructure and MR is going to be the provider.
- » Uh so the big vision has been evolving and as you stated we started uh as a company that built models uh because with Arthur and Guom this was what we knew how to do at the start.
- The premise on which we built Misual AI was immediately solving for enterprise needs uh and we started with open weights model.
- After this uh and working with enterprise we realized uh the need for basically the rest of the stack.
- So we built uh the serving platform because infrastructure was needed.
- Um and then all of the tooling around it uh was also something that we saw was missing.
- more than the tooling, it also requires a lot of work and expertise still to get deep into uh an enterprise workflows and really help that transformation.
- And so we built that uh FDE function and more recently uh with MSR compute uh we’re going a bit lower uh in the stack as well.
- So we’ve done all of this uh because it was required for enterprise success uh while still continuing uh on our models journey.
- All of this stack uh being modular is really important to us as it gives full control to u enterprise and our clients as to which part of the stack they decide to uh own and control which is maybe more involved or that they decide to have serverless or basically this modularity that we like.
- » All right.
Building Own Data Centers
- So let’s take some of those modular components uh in in order.
- Let’s start with mistral compute.
- So that was a big announcement uh I guess in June of 2025 putting a big partnership with Nvidia to um help with this effort.
- Uh what’s the current status?
- Is that live yet?
- Are you building it?
- You know, how does one go about building data centers or or leveraging data centers in Europe?
- » Maybe first to go into the reasons uh why we decided to start building our own data centers.
- uh we tried uh a lot of different partners over the years and we realized that our use uh of the AI compute for large scale training was not necessarily well understood by a lot of providers and our uh need for stability especially like when you run inference on a few GPUs or when you run small scale trainings on a hundreds of GPUs margin for error is a lot larger than when you run trainings on uh thousands of GPUs at the same time.
- And so to address this need for stability, we saw a way for us to basically build our own data centers and maintain it with our understanding of what quality looks like.
- And so that was why we uh launched MRL compute.
- And when we decided to do it, we also realized well maybe others will benefit from it.
- We launched into uh a bigger uh basically development than what was previously intended.
- And so this was announced in June as you said since then the building of the facility has progressed quite well.
- It’s in the south of Paris and we are right now running through the stabilization uh stabilization of the first trench.
- Uh so it’s uh quite a large data center so delivery doesn’t happen in one day.
- And the first part of this data center is something that we are working on as we speak.
- We have a few jobs running and we’re fine-tuning uh basically all of the last uh things uh to run at speed and with the right stability.
Compute as a Service
- » Okay, great.
- And uh did I understand correctly, it’s going to be for your customers and your own needs uh around training, but also you’ll be providing it as a service to others uh in in Europe and beyond.
- » Yeah, exactly.
- So we will use part of that capacity for ourselves as one of our training clusters but we will also provide a managed Kubernetes and managed serum stack on top.
- » Okay.
Lessons Learned in Data Center Construction
- Any uh lessons learned so far?
- I mean as you said you guys come from a very deep background in in AI and AI research.
- It’s a whole different thing to build a whole like data center facility.
- How have you gone about it and uh what are some things that that surprised you and any lessons so far?
- As most uh new experiences as a founder, I relied on the knowledge of others.
- Uh and so I was uh lucky to have a very a few seasoned uh HPC experts uh and and a lot of uh cloud software experts as well to build that solution.
- For me personally, and it’s one of the things I love about uh my position at Mistral is that I get to uh discover so many new things uh and so many new problems I hadn’t thought possible.
- having to learn to like all of the different parts of building a data center, all of the different trades that you have to coordinate, uh all of the potential um synchronization uh between all of the different trades.
- I mean, it’s a huge building.
- It involves hundreds of people working on it.
- You have this then when you uh stand up the thing, uh you have to question what works.
- You have to filter through the blades that are faulty.
- It’s just an entire new area of work where I get to see um experts in their field go through things and try to explain to me what their daily work is.
- It’s always fascinating to see um an expert in his field like do something that you don’t know how to do.
- I think the logistics of it uh and the timelines are also quite different from what I’m usually um dealing with in software and research.
- for new capacity to uh be built, you have to plan around uh having energy available, you have to plan for the uh space to be available and on time.
- And so it’s a lot more long-term planning than a few software features.
Energy Considerations in Europe
- » How do you guys go about power since you mentioned energy?
- » In what we’ve been doing in Europe so far hasn’t been a huge blocker, although uh there is constraint.
- Uh I think the grid in various parts of Europe is not uh necessarily easily extensible.
- I know it’s uh an issue in in France.
- A lot of the sites are uh contended.
- Um so we we’ll see how it all develops.
- We are lucky in Europe to have uh very uh clean and affordable energy uh either with uh green energy in the Nordics and nuclear in France.
- So it’s it’s been relatively okay for us today.
- as you describe this uh what comes to mind is the gigantic amounts of money that are being invested in the US around uh data centers.
Competing with Large Tech Players
- How do you guys uh go about that from a financing standpoint and and perhaps even more taking a step back if you think about the race between the big AI labs globally whether that’s you know the opensis and anthropic of the world and and XAI uh it seems that all of them are affiliated with a gigantic pocket of money somewhere obviously there’s Gemini and Google to add to the list and and Meta I’m just curious like how where do you guys stand on on that you have a bunch of partnerships with
- um SAP and Nvidia but there is you don’t have one of those gigantic companies on your cap table.
- So how do you how do you think about uh competing in that general context?
- So with those uh companies, so the hypers scalers, it’s um there are two parts to the game and we’ve played the partnership uh part quite well with them and we're integrated within Google's Verex uh Amazon Bedrock and uh Asia Studio and that is uh the choice that we’ve made in term of having access to uh gigantic pockets of monies.
- We’ve been focused on efficiency from the start.
- Uh and I think we’ve done quite well at building uh models that are uh competitive with the uh investments that we’ve uh put in.
- For us, it’s important to build uh the company as efficiently as we can.
- uh and I deeply believe that with the capabilities that we have today in the models there is so much to be unlocked uh in enterprise that um I I don’t think my main focus uh today would be into going into the gigawatts of power we still need to build uh so much with our clients and unlock so much values with theap capacities that we have » all right so let’s go into uh the enterprise reality of all of this um so if I’m an enterprise or if I’m a sovereign and I want to deploy a MR
Enterprise Deployment and Customization
- open-source model what is it that I do these days with everything that you that you’ve built the way we work with um enterprise I mean as you mentioned like we have a few of our models that are open source and Apache and all of our clients are welcome to use them uh as they need what we have seen in terms of success is that given the current stack It still requires um a lot of expertise uh to manage to come to um actual value and um and things that go to production.
- Basically the way we interact is that we usually stand up our um Misual AI studio which is our platform and we can deploy uh all of our stack on the client’s choice uh of deployment methods.
- So it can be on prem uh it can be on their VPC it can be on uh in several places.
- The reason we do this is that it lets uh clients build where their data is uh and without having to shuffle things around which as I’ve learned as a CTO is something that you don’t want to do ever because it asks it raises a lot of questions uh and it’s uh quite a stressful thing to do.
- So once this is deployed uh we then uh work with the business units to understand where their pain points are.
- Sometimes it's knowledge management and I think it's the most well-known uh use case from the output from the outside of the enterprise world but it's also around um automating core workflows for the enterprise.
- Um it’s you know some tooling that you wouldn’t expect where one thing that we’ve done is around code modernization uh where you you turn a bunch of Excel sheets into an actual like Python app.
- Uh and if you have many many of those sheets then potentially you want to use AI for this.
- So once the infrastructure is built then we basically look for what’s the most valuable to the customer and we start acrewing value uh inside a stack of AI assets that then accelerates all of the other developments with that customer » and is part of the idea is that you do actual model work at the customer and for the customers in particular fine-tuning.
- » Yes, we we customize in various ways.
- Uh so we have done continued pre-training and this is most useful when you want to uh change the capabilities of a model uh more deeply.
- So we’ve done this to sometimes change uh the mix of languages in a model to get something that’s a lot better at thou east Asian uh languages for example or you could have require this if your internal data uh which doesn’t happen on the public web is something that’s so new uh that you need a large amount of to of tokens uh to get a model that understands it and becomes fluent with it.
- So we do uh these kinds of continued pre-training fine-tuning.
- We also um like and this is more for an efficiency reason.
- When you get to smaller models, you have to make trade-offs.
- Uh the models won’t be as good in their knowledge of the world.
- And so when you lose uh a lot of things, you have to focus on what you really care about.
- And so this is typically important if you want really uh fast, really cheap uh models that will be really good at a specific task.
- It’s also useful if you want models that run on the edge uh that get very very tiny.
- Uh and so for all of these fine-tuning is a tool of choice.
- Another uh reason to do fine-tuning.
- It can be to adapt to uh data that’s not necessarily massive but that’s also not available on the web.
- So typically in coding uh what happens is that you will have massive code bases sometimes acrewed over decades uh that the model will need to be able to uh work with in terms of uh having like vibe uh deployed on it typically and so being able to come in not move the code base and uh learn an actual coding agent for that codebase is really powerful as well.
- » And who does the all of this you have evolved towards an FDA model.
- So we have indeed a large uh FD section.
- It’s it’s a mix of software and uh FDEEs and we split our FDES into what we called um AI engineers and applied scientists.
- Um and so uh applied scientists will tend to use the tools that we’ve just uh uh talked about.
- So fine-tuning, continued pre-training and the likes where AI engineers will focus more on adaptation to the enterprise environment and figuring out what workflows to automate and all of this.
- They work with the customers to make sure uh that the use cases are indeed providing values and going to production.
- But it’s also a fantastic way for us to understand what matters in an enterprise context and be faster at building the right platform.
- And uh again those customers are the kind of customer for whom customization and privacy is essential.
- Uh how do you how do you position again open of the world that are going very hard at the enterprise?
- Is that data sovereignty?
- Is that customization?
Defining Enterprise Control
- » The term we use is control.
- The value that we see is both in our expertise and the software stack that we provide.
- The software stack once deployed uh is in the hands of our customers and they can change it, they can add to it.
- They own model changes that we make and I think it’s really important as a customer to And so in working with us and building uh because it takes effort uh to build an AI advantage uh today and so having this effort built into uh something that you own is I think a choice that makes sense.
Agents as Building Blocks
- » Let’s talk about uh agents uh obviously part of the overall effort at Mistrol.
- How does that work?
- Uh how do you uh build an agent and uh what key use cases have you seen so far?
- Personally, I think I've moved uh from agents to uh workflows, which is I guess an abstraction uh on top.
- Um so agents are I think the building blocks uh where you have a given expected input, a set of tools and you are trying to reach a uh set of uh you have a goal that you want to reach.
- The set of inputs uh that we’ve enabled are um images, text uh and audio.
- When you build an agent, to me it’s really important that you build it on a focused uh task with a data set that you understand and that you can iterate on and that you can improve.
- What we see in enterprise is rarely things that are solved with agents because that's not necessarily where you would expect uh an FDE to be most useful.
- Those ideally would be built uh on our platform by the customers directly.
- Where there is more values value is in uh more complex workflows where you will have several uh agents interact through a workflow to automate something slightly more complex.
- And so that’s what we’ve been focusing on.
Automating Complex Workflows: CGM Example
- What would be an example?
- » An example is something that we’ve built uh with the shipping company CMACGM where we’ve automated the uh container release process.
- Uh and so it’s um a use case where I I don’t know how familiar you are with shipping.
- I wasn’t at first.
- Uh but a container reaches a port and you have to uh harbor uh probably in English.
- some decision has to be made that this uh container is ready for release to the uh next person on on the line to handle this container and so there are lots of uh checks uh that need to be um run and data to be accessed in the back end uh before that decision is made.
- So as you can imagine, some of those containers are extremely valuable and you can’t really afford a mistake.
- And so what we’ve done in this case is an application that’s integrated into um how these uh harbor worker work and it automates a lot of the manual work that they did to check the data and they make the final decision uh given all of the evidence.
- » Okay, this is super interesting.
Trust and Governance in Agents
- Obviously the the key question about agents these days especially when they are combined into workflows is the question of uh autonomy.
- How do you guys think about it?
- How autonomous are those agents uh in in your deployments?
- » I don’t know if it’s the way I think about it.
- To me the better question usually is how much you trust uh the agents and there are a few dimensions uh around this.
- What worries me when building those kind of workflows is that typically if you want the value to acrue and if you want to build faster and faster the more workflows that you build, what you will want to do is uh reuse assets and make them reusable by others.
- Uh as soon as you do this with agents, you then start to ask the question, well this agent has access to some data that is privileged uh but maybe this other agent uh is publishing it to something that’s public.
- You might have governance concerns where uh some agent is acting on something very critical and you don't know necessarily that the data that it got uh has been approved or something like this.
- It’s really a new way to develop where uh the parts of your workflows have to be trusted.
- Each of them to be trusted requires uh quite a lot of tooling uh and quite a lot of observability uh to get confidence and to basically enable this at scale in an enterprise.
- So the question that you’re asking about autonomy to me this is something that I see happening when I vibe code.
- Sure like longer running tasks and making and improving on this is going to be critical and we’re uh working on it daily.
- But today, the problems that we’re solving on the software side of things are really about how you trust what you’ve built and how you improve it.
- Uh, and how you allow an entire company to build on it with confidence.
Studio Components and Versioning
- » Maybe describe some of the things that you guys have built in studio around governance as you mentioned and trackability and uh registry all the things.
- What what are the key components of an a modern agent suite?
- So workflows as I mentioned is something that uh we’ve worked a lot on uh with our customers and it’s not GA yet.
- Uh so look out for this uh sometimes in the future but it’s also one of the benefits uh of working with enterprise we can um have a lot of design partners and once we’re confident uh with the solution uh we we make it G. So a workflow solution is critical.
- Workflows are built on various uh model capabilities.
- So u vision, audio and text and reasoning.
- It is important to uh have a registry of uh connectors and MCPS.
- Uh and so for this we have uh our connections.
- The observability is an area where we’re still working on.
- Um it’s important for me to be able to iterate and really define uh precisely what an agent does and control each of its goal uh and see how it’s progressing um being able to maintain evaluations and uh build build on them.
- What is um difficult in this entire sea of complexity is that you also have to maintain proper versioning and tagging and think about how you’re going to deploy and improve uh upon what you’ve built.
- So let’s say you’ve built a kickass workflow based on a lot of agents and models that Mrol has released in the past.
- Then a few months pass and there are new sets of models that are out.
- Maybe you can simplify that workflow.
- Maybe the next uh mist 4 is good enough that you can factor out a few agents.
- Basically, what you need to be able to do is create a new agent, run it on the same set of inputs and outputs and control that you haven’t broken anything and then deploy it in the wild.
Context Graph and Enterprise Context
- All of this software uh suite basically which has been built for software development over years I feel it isn’t there yet uh in the AI world and that’s what we’re building » as I’m sure you’ve seen there was uh for the last few weeks in startup and venture circles there’s been this whole idea of the context graph as an infrastructure that made the rounds.
- Is that something that you think about or a layer that would basically uh enable one to know how the agents made a decision and how those decision relate to one another?
- » I’ve seen this indeed and I think there are two uh levels to that discussion.
- the part that you mentioned at the end where uh it’s interesting to know how an agent came to so in that discussion when when we talk about understanding how an agent came to a decision or an action the game is really to understand how a human uh agent really made this decision.
- It’s understanding how an enterprise does what it does and it’s certainly interesting.
- uh what keeps me up at night and what I really want to solve first is just the basic idea of gathering a workable enterprise context.
- Right now uh with uh any model uh and with a lot of effort you will be able to get some connections to tools and you will ask a questions and your agent will do a bunch of things.
- it will realize that oh by doing five API calls and three joins I can probably get uh what Timothy asked immediately what should happen is that um all of that uh discovery and all of that intelligence should be stored somewhere to be reused.
- It’s not really how things happen.
- It’s just basic knowledge uh about what the infrastructure of the company is.
- So knowing where the tables are, what they contain, how they’re joined.
- So all of this um is compute that should be amortized basically and to me it's really the entire game with the context engine as we call it internally is to um be in a setup where over time knowledge of the company and the context that's available to the agent uh acrru and is maintained.
- The second order thing of oh how was that decision reached?
- Sure.
- Uh it’s going to be super interesting and it’s important, but right now I feel we’re not even in a place where it’s easy for an enterprise to have any worker uh in it be able to build an agent that has access to the right context.
- For this to happen, you have huge uh data privacy concern.
- If you want this to be efficient, you need to give access to uh the agent system to the entire uh data of your enterprise.
- And there is going to be arbbacks everywhere and you need to make this safe.
Current Reality of Enterprise Deployments
- » Speaking of which, what what’s current reality of enterprise deployments of of generative AI from your perspective?
- just listening to like some of the concern like since like we very early » to me we are still in the building phase and I think it’s kind of the frustrating thing for enterprise is that when you come to um a chat assistant you feel that it’s it’s magic and it’s all going to work but as most things that have value in life there is still work to be done uh to get to them and so most of the enterprise value of AI will happen once you’ve gone through that first building phase of just setting up
- all of the machinery.
- You’ve got to set up all of the connections.
- You’ve got to make all of that data available.
- And the reality is even despite a lot of work recently to make u data more available in enterprise, it’s still not easily available in the format and at the scale that we need uh for the true ROI of AI to to happen.
- And so when we come in uh there is still that phase of work that is uh just work uh to connect everything and then be able to build on it.
- » So do you think we are years away from generi actually being deployed in the enterprise?
- not years uh I think years singular uh it’s uh also to be fair to us we’ve started working I mean the company started two years ago and so most of our uh » it’s a good reminder right it’s a good reminder that like you guys have have done all of this and the company was started in yeah June 23 right if I recall » yeah and so for most of our clients uh we we started working with them recently the tooling uh for everyone is still in its infancy And so I hope that the tooling will stabil
- stabilize uh and I hope that we will have true value.
- True value to me is really okay we’ve gone through that first phase of building connections and now employees of that enterprise are able to use everything that we’ve built.
- Right now I think we’re in a phase where we build siloed things uh because we’re scared of uh data going through walls and everything.
- And so to me, the real success is when you’re confident enough to give all of that control back to the company’s employees at large and they start really building on it.
Future Enterprise Demand Growth
- » You’re talking about MRO in particular about the industry in general, right?
- Is that do I understand this correctly?
- Uh because obviously that’s that’s the big question, right?
- we we all collectively building this whole thing and data centers and models and pouring uh billions and I think it’s pretty clear that from a personal use case or uh from maybe some discrete like coding use cases like the the demand is very clear uh but the big question is whether demand is going to materialize at the same level as the extraordinary level of supply we’re building » yeah around this I think the expectation is that demand and basically amount of tokens generated uh for the
- enterprise will uh completely jump once you are not bound anymore by humans asking questions or reading them.
- As soon as you have enough trust uh to have agents running in the background, as soon as you’ve set them to run a bunch of ETLs, as you’ve got them running lots of workloads, uh and you’ve got them consolidating data and knowledge across your entire company, then you’re not really um limited by the number of tokens that humans can create or read.
- And so we I think everyone in the industry expect the demand to jump at that point.
- And the reality is for this to happen, you just need a lot of boring software and control and things like this.
- » It’s amazing how much uh all of this is engineering, right?
Engineering vs. Model Performance
- Versus just sheer performance of uh of models.
- » Yeah, it’s a lot of plumbing and the goal is to make all of this plumbing easy and easier and to make it faster.
- » All right.
- And you said we’re about a year away.
- I » I’m not the most optimistic person.
- It might be faster.
- Uh who knows?
Banger Use Cases and ROI Drivers
- And we we talked about use cases a bit already, but let’s just put that one to to bed because it’s such an important question.
- What do you think are the kind of the banger uh use cases in the enterprise?
- Let’s assume like all agents work uh in in a in a workflow kind of way that you describe uh based on either your uh industry watch or or more specifically talking to your customers.
- What is it that is going to generate a amazing ROI beyond coding which is pretty established at this at this stage?
- » Yeah, there are several dimensions to this.
- Coding is an obvious one and um to me to get the full um ROI of coding you need customization.
- Uh because a lot of ROI is unlocked uh on like sprawling code bases that are completely impossible to know for uh for something that’s been trained on the web.
- uh if you’ve got uh an enterprise that’s been building its own like domain specific languages for years, you’ll need some customization for an agent to come in and be competent uh in that respect.
- Um so coding is definitely a big one.
- Um if everything uh comes true as I hope I think there is still a huge jump in how we accelerate knowledge worker um and I believe the magical experience of uh you go to your chat assistant it’s connected to your system and you can ask it anything uh about the enterprise just hasn’t realized yet and it’s really obvious uh when you see the kind of queries that people are making expecting them to just work.
- And to me who’s building the system, it it feels like magic.
- Like if you need to somehow send an email to three people and coordinate a meeting and also like gather data from some BI system, it’s just something that requires um a lot more plumbing and capabilities that we have today.
- Um so that’s going to be a huge lift.
- And I think the last one which is maybe closer to my heart is really when we start to customize models to uh a kind of data that is particular to an industry.
- So typically if we uh work in oil and gas they will have systemic data that we can help uh understand and make sense of.
- If we work with um computer assisted designs, uh they might have uh full databases of specific data formats that are not widely understood by the most general models yet.
- And if we manage to build a system where in a light touch way from us or in in my dream world, we don’t really have to intervene.
- It’s all uh self-s served for the customers.
- they can consolidate that data and then build themselves a model that really understands what their actual uh private IP is made of and make sense of this.
- Uh then I’ll be super happy and I think there is huge value to unlock there.
- » Great.
Reasons for Edge Deployment
- Where does the edge uh fit in all of this?
- » There are a few reasons to go edge.
- Uh first there are some regions where it’s more convenient to um be able to work without internet and there are also a lot of capabilities that don’t necessarily require uh a huge model.
- So if you just need something that goes uh voice to action on any device uh today with uh typically the voxal models that we develop this is doable.
- Again, an area where the more uh focused your use case is, the smaller you can make the model through fine-tuning or um through just distillation in a in an even smaller architecture.
- I think voice to uh action is going to be a big use case.
- I think it will simplify a lot uh the current stacks uh for these types of things.
- There is also some privacy things uh where you could imagine uh all of the context consolidation stays on your personal device and for most things uh you can deal with a small model uh that answers a lot of your questions and then you potentially can gate uh what goes out to uh another like cloud-based models.
- I myself take the train a lot.
- Uh I like having coding assistance.
Defense Industry Applications
- uh having uh DevTool run on my laptop while I code on the train is uh comfortable despite the bad Wi-Fi » and uh presumably there are some uh defense uh use cases as well.
- So you you guys do quite a bit of defense work as I understand it with France with with Germany.
- I think you you mentioned some partnership with Helsing is AI on drones and that kind of stuff.
- Is that a reality?
- » A reality it’s uh it’s something that we work on.
- Yes, we have a robotics division that works with these uh partners.
- Having a very um well- definfined use cases uh makes us able to really take the model down to u lighter uh types of sizes.
- Um and it’s of course uh use cases where control is super critical uh and you need to be um yeah able to really validate the solution.
- » All right, let’s switch to the model part of the discussion.
Mistral 3 and MoE Architecture
- In December, you guys uh released Mistrol 3, which was a big release still with thee architecture, which is at the core of what you guys have been um doing.
- You mentioned efficiency uh earlier in the conversation.
- maybe walk us through the general thinking and and approach like in a highly competitive world uh of uh AI models both in terms of closed source but also very much open source and all the Chinese labs.
- What is it that you guys are trying to do and how do you position?
- Yeah.
- So we’ve released Mistral Large 3 which is uh an MOE.
- MOEs are uh really nice systems to train uh because of the lower uh amount of flops which uh makes us able to push performances um a lot more uh during training.
- They are not necessarily the best formats for uh on-prem deployment because as of today uh if you want to get uh the best efficiency out of uh a mixture of experts model you require a lot of volume uh because you’re looking at deployments across dozens of GPUs usually um and to justify that amount of GPUs uh you need to have the right throughput.
- We are training uh large moes to get the best performance um with the most efficiency during training.
- We are also continuing to train u dense models at other scales because depending on the environments uh in which our clients want to deploy this might be the more uh costefficient solution.
- I think both architectures are still valuable.
- um on edge as well.
- Uh sometimes you just don’t have the RAM capacity uh to deploy something like a sparse mixture of experts and so going dense is helpful there as well.
- But yeah, definitely for training uh mixture of experts and their lower flops are very interesting.
Ultimate Goal of Model Development
- » What is the ultimate goal of the model effort?
- I mean clearly you guys are a frontier AI lab but um are you trying to create the the best models and and solve AGI or are you trying to be the best open-source model compared to the Chinese labs or you know whatever open source eventually comes out of the uh US what is it that you’re trying to do » we’re trying to get the best uh models that we can and the model that’s most useful for uh the use cases that we cover uh in enterprise.
- And so typically with the rise of uh agentic uh behavior, one thing that’s very important is how you deal with uh various contexts, how you deal with various um documents uh being added to the input.
- And so having the capabilities to do architecture iterations really trying new things in terms of model training is critical.
- Um so we’re pushing the boundaries of what the current models can do with uh the compute capacity that we have but we’re also trying to focus on the things that are is most annoying uh in our deployments today.
- And so one of the consideration that has been solved with a few harness uh tricks is the context of uh those agentic systems.
- So it’s visible typically in vibe coding but it’s um definitely uh applicable to a lot of other use cases where through all of the tool calls you’ll have to uh consolidate uh and summarize the context to be able to fit everything and uh have the model focus on the right parts.
- To me this is just an artifact of the current architectures.
Context Window Limitations and Solutions
- uh we’re trying to fit uh things in a linear context windows where essentially the questions that we’re asking aren’t really necessarily all linear.
- Um and so we rely today on the file system for this and that I think that was the big change in u and realization through vibe coding is that agents are good enough at uh manipulating file systems that they can use this as a replacement for uh their context window.
- Basically uh they can select parts of what they want to read.
- they can select parts of the tool results uh and this minimizes uh the context length requirements.
- This is the state today.
- I think we can do much better and I think there is a lot of uh improvements to be done on those types of uh questions.
- » Do your agents run on sandboxes?
Agent Sandboxing and Isolation
- » It depends on the types of agents.
- Uh but the answer would be yes.
- If it’s uh if it’s coding agents, usually uh we have uh sandboxes that will let the agent iterate uh and run.
- I think the depth of the uh isolation will depend on the use case.
- Uh typically if the file system is just representing textual context and you’re not expecting the agent to do much action on it, then you don’t really need a full sandbox.
- Uh you just need some representation of that context as a file system and it can be any sort of abstraction.
- But if you are I don’t know typically running asynchronous code development then yes you need a sandbox.
- » Great.
Constraints for Mistral 4
- What is the current constraint that um you guys are facing to make uh MR 4 when it eventually comes out do much better than ML 3.
Future Model Development: Compute, Data, and Synthetic Data
- Is that a question of MR compute or is that a question of of data and uh in particular are you guys doing anything around synthetic data that you can talk about?
- Definitely compute and uh the current deployment that we have will help uh as it’s going to be giving us a lot more grace blackwell capacity than we had in the past.
- And so that’s uh something that we’re very excited about.
- And when you add uh compute, you also have to add data.
- And so we’ve been hard at work uh making sure that our uh data mixtures are uh as high quality as ever and growing in size.
- But as you mentioned, one of the ways to do this is through synthetic synthetic data.
- In terms of um where we use synthetic data the most.
- I think a lot of the interesting work that’s happening is for the post-training part where we can um build environments uh that look similar to uh an enterprise and then uh try to uh synthetically create queries that are hard and that will require multiple hops.
- And so all of this work um is in addition to the coding work, the reasoning work is really what makes the final model able to perform uh in the various uh environments that we work in.
- So before it was about uh acrewing world knowledge and the uh web helps a lot with this.
- Now it’s more and more about acquiring knowhow.
- Uh and for this uh it’s really about um trying to find what our uh customers are trying to do, trying to replicate it inside of our training environment and uh let the the model run basically.
Pre-training vs. Post-training and Reinforcement Learning
- » You mentioned post-training and that’s one of the key topics of the last 12 months in particular this evolution of um LMS uh into systems with both pre-training and post- training and a lot of reinforcement learning.
- Where do you guys uh fall in that spectrum?
- Are you uh pushing a lot of uh reinforcement learning?
- Do you believe that pre-training has still room to grow?
- How do you think about it?
- » Yeah, everything still has room to grow.
- What I’m interested in as the CTO is really how you make uh all of the steps of the pipeline uh work well together and how everyone can uh develop most efficiently.
- Um, typically what happens in uh post training is that you will have a team that’s working on uh improving code.
- You will have another team that’s improving um different uh enterprise uh behaviors.
- You will have another team that’s uh improving on uh instruction following.
- Uh and so all of this uh at some point has to come together because customers aren’t happy if you require them to deploy five different uh models to get their job done.
- There is really an internal engine and capability around making all of these work stream come together uh in the way that you expect that is super interesting to build and so but yeah uh internally we’re building and improving all of the parts of the stack.
- I think the post training is very rich because it also touches all of the new use cases of LLMs and I think it’s been very exciting to see just all of the the new use case that pop up every day.
- Anytime someone on Twitter finds a new exciting things that they’ve done, then suddenly, you know, you’ve got to make it this proof of concept into potentially a base capability on which your model will perform well.
- And that’s uh potentially an entire stream of work and you’ve got to do this efficiently and prioritize.
- Well, » where doesing fall in all of this?
Reasoning and Tool Usage Integration
- Uh you guys launched a reasoning model called Magistro a few months ago.
- Is that is that a big priority?
- So reasoning is a a big priority.
- And the interesting thing about reasoning was really how you can train models with reinforcement learning.
- And so it was first shown through reasoning, uh, because the system would learn to create better reasoning traces to, uh, get to better results.
- But the system is the same whether you create reasoning traces, or whether you iterate on the tools that you call, or mix them both.
- And so I think more and more the way to train, uh, all of this, uh, is going to come together.
- And sometimes you’ll have reasoning traces, sometimes they’ll be long, sometimes they’ll be short, sometimes there won’t be any because it’s not necessary.
- And there’s no real difference between creating a new thinking trace or calling the right tool.
- It’s, it’s all the same to me because what you’re optimizing at the end is what is the best, uh, output for the model to create before it gets a results, to, uh, to me.
- Great.
Devstrol and Vibe CLI: Agentic Coding and Enterprise Intelligence
- Let’s talk about, uh, Devstrol 2 and the Vibe CLI.
- So walk us through those products and what they do and, uh, why people should use them.
- Sure.
- Um, so DevTool is our, uh, agent tech coding model.
- And so it’s something that you typically vibe code with.
- And you are more than welcome to vibe code with it through our CLI aptly named Vibe.
- Value of vibe coding and why we focus on it.
- Coding is a huge use case in enterprise, um, and especially a lot of our clients have, uh, yeah, large code databases where it’s helpful for us, uh, to take our system and customize it to their codebase to let, um, our agent run.
- Now, the Devstrol and agentic coding is not only about, uh, vibe coding.
- The same system when you run it, uh, asynchronously can be used to review PRs.
- Uh, it can be used to check code for specific conditions.
- It can be used to modernize code.
- So its applications even in coding are, uh, quite wide, as I alluded to as well.
- Um, having a system that, uh, is good at handling a file system is more generally very interesting.
- Uh, even if you’re not using it to code, you can use it to reason, uh, about enterprise knowledge.
- You can use it to connect to enterprise systems, and it’s, to me, it’s the basis of really the enterprise intelligence that we’re starting to build.
- And so the big news is, yeah, the, that those systems are, uh, going GA.
- Uh, we’ve got, uh, an offer where chat users, so Luca, our assistant, um, will also, uh, get the ability to use Vibe and the associated models, and we’re trying to, uh, basically make that usage as wide as possible.
OCR3 and Document Processing
- Another thing that you, uh, released reasonably recently, I believe, is, uh, OCR3.
- What does that do?
- That enables you to just like, uh, scan any, uh, any form, any document.
- Yeah, OCR is a huge use case in enterprise.
- Uh, a lot of our customers have, I mean, the typical example is KYC where someone will submit a form and you need to input that information in a structured way in your systems or you need to reason about it.
- And so OCR, interestingly, is, uh, it’s not the types of systems that I would have expected, uh, LLM to really, uh, make large strides on.
- The visual reasoning and the visual understanding has gotten so good that it’s, it’s just an easier way to process things.
- Uh, in my mind, you have any sort of input, um, and you can get the the data that you care about.
- As I mentioned, when you build agents, you have, uh, a different type of inputs for the task that you’re trying to solve.
- Documents and visual information are just a very, very frequent kind of kind of input.
- Uh, sometimes it’s a lot cheaper, uh, to use a small OCR model to just get the text that you care about and then potentially post-process it or deal with it with another system than to run it through a large, uh, multimodal model that will, you, basically do the same thing but at a higher cost.
Multimodality: Image, Audio, and Video
- Yeah, you mentioned multimodal.
- To to which extent is Mistrol multimodal, or to which extent is that, um, voice is is video, something that you guys either do or think about, or is that just not a big enterprise use case?
- So to answer on the first part of the question on whether, uh, we build multimodal models.
- Yes.
- uh it’s always a balance between exploring in a direction, getting good capabilities and getting the first model out there and then integrating it uh into the trunk like the main model that we use for everything else.
- And so those will always happen at separate times but for uh audio um we have uh voxil as I mentioned and all of our um main models uh understand images and can reason about them.
- for videos.
- It’s a subject that we tackle through the lens of robotics uh first and so we’re doing our first explorations on that topic.
- » Okay.
- Well, again the the velocity uh has been super interesting to uh to to watch.
- I um again appreciate you your reminding us that you guys have been doing this for only uh a couple of years.
- So um just uh very impressive all together.
Engineering Efficiency and Team Building
- maybe take taking a step back and thinking all of this in terms of uh engineering and lessons for for for builders.
- So as we alluded to a couple of times through the conversation like you you you guys are doing a lot with uh comparatively it’s always it’s very relative in the world of AI less uh resources.
- How have you uh been able to do this from an efficiency standpoint?
- We focused on the parts that we knew would provide the most uh impact uh and we focused on basically what we could afford at different times.
- So when we started and we had uh enough resources to train uh a few models and uh then we focused on getting the data perfect because we knew um this was potentially not the most exciting part of the work but it was absolutely critical and any improvement uh on the data quality would 10x the uh improvements that we would get by really um improving on the model architecture or things like this.
- And so I think it’s focusing the right effort uh depending on the scale and the um yeah depending on the scale of the company » and from a team uh building perspective how have you gone about it the the three of you the three co-founders have a deep background in in AI um are you these day focused mostly on building like an FDA team or are you still uh building this large kind of like research lab effort and how do you uh think about the right ratio?
- » We are growing uh all of our teams both uh research uh FDES uh product engineering uh infrastructure for compute and all of the teams have their own uh challenges in how you build and what order you uh recruit people in.
- It’s been important to me um at the start to I mean to me and uh and GM and Arthur we both like the three of us were uh good AI practitioners so we knew how to train models and we knew how to code and so we started with people like us to get to the models trained the fastest um but that doesn’t work as you scale uh you it is critical to build the right uh infrastructure uh for research And so this takes different skill sets.
- Uh and it’s something that we’ve been uh building over the years as well.
- Uh and it’s fascinating as someone who used to do uh research in a at a smaller scale to see the kind of systems that are involved and the the gains uh that you can have at scale.
- Uh in terms of engineering, it’s kind of the same story really.
- uh where you start with um a team that’s broad in its knowledge and self-sufficient and can iterate fast and then more and more you bring in experts or people that are that have seen larger scale and will tell you like well this won’t work in six months and so we should fix that now.
- So, it’s been super interesting growing the company and seeing all of the uh successive things that break at each scale and overcoming them through either changing the system, changing the organization or building new things.
- » How have you navigated the whole Europe to US and rest of the world dimension of this?
Global Operations and Company Philosophy
- I you’re the very much the the pride of France, the pride of uh Europe as well equally.
- This is a global race.
- How have you uh made it work?
- » So, we work um on all three continents.
- We have offices uh in PaloAlto.
- We have offices in Singapore as well.
- Most of our employees work uh from Paris.
- It’s a good representation of uh what we’re trying to build, which is a solution that’s uh independent and that people control.
- and in and this target uh it doesn’t really matter uh where we’re from or who we’re building for.
- Uh we provide the tools uh and the customer the end customer then owns uh everything that’s built on it.
- And so I I think it it hasn’t really been something that I’ve spent much thought on.
Future Outlook: ROI and Democratization
- » So uh what what should we um expect from uh Mistl over the next uh couple of years?
- Over the next couple of years, I would say uh diminishing doubts on the ROI of AI uh ideally so faster uh time to success uh larger and larger uh use cases being built and really democratization uh of building tools with AI in enterprise.
- I think this is really what I target for our customers.
- uh it should be easy uh and most people should be able to accelerate themselves through the use of AI.
- I think we’ve seen this happen quite uh impressively for coding and it should be something that happens uh a lot more widely.
- » I was uh struck throughout this uh conversation by how pragmatic uh you you are and and focused on precise goals around enterprise success.
AGI and Enterprise Control
- What do you make uh of the whole, you know, rush to AGI conversation and people being AGI pill in San Francisco and other places?
- Is that is that something that you see happening or does that to some extent not matter from your perspective?
- » I mean it it matters because the the better your systems are, the more uh impressive things you’ll be able to do and it it’ll become easier and easier.
- requirements I see for control and governance in enterprise make me think that even if I had uh some AGIS model on my uh servers right now if I were to go uh into a large bank and say here is a thing please let it control everything for you they wouldn’t be happy to let it do it and so I think building the infrastructure uh properly is uh quite key to following the progress of these models and really being able to quickly unleash all of their capabilities.
- So to me it’s it’s two directions that are necessary.
- You need to improve the capabilities of the model and it’s super exciting to do so but the journey of uh making it trivial and uh easy for everyone to unleash those models on your enterprise workflows uh without really wondering what’s going to happen is is equally important.
- And honestly super uh super fun as well to develop.
Concluding Remarks and Appreciation
- There are lots of super interesting questions.
- » Wonderful.
- Well, Timote, thank you so much for uh doing this uh deep dive on Mistrol with us.
- It’s been fascinating.
- Congratulations on everything that you’ve built again in this very short period of time.
- Uh and excited for what’s uh coming next.
- So, thank you for spending time with us.
Podcast Outro and Listener Engagement
- » Thanks.
- It was a pleasure.
- » Hi, it’s Matt Kirk again.
- Thanks for listening to this episode of the Mad Podcast.
- If you enjoyed it, we’d be very grateful if you would consider subscribing if you haven’t already, or leaving a positive review or comment on whichever platform you’re watching this or listening to this episode from.
- This really helps us build a podcast and get great guests.
- Thanks, and see you at the next episode.