Oct 11, 2023 · 52m · mad
Secure, Private, Powerful: Dust’s Vision for Enterprise AI Agents | Stanislas Polu
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The MAD Podcast, host Matt Turck interviews Stanislas Polu, Co-Founder and CEO of Dust, exploring his career at Stripe and OpenAI, the product design and RAG architecture behind Dust's enterprise AI platform, and the rapid growth of the European AI ecosystem.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 18.6% of the talking time here. How this is scored →
Matt as informed peer 3.2 Guest teaching 4.3 Guest disagreement 0.5 Matt pushing back 0.3
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 5:53 Dismissing the bidding war label
Stanislas explicitly corrects Matt's characterization of his startup exit, stating 'I wouldn't call that a bidding war' and clarifying that Pinterest made an offer before Stripe was contacted.
Hardest push from Matt ▶ 31:13 Challenging product vs services business modelMatt directly challenges Stanislas on whether Dust is effectively operating as a services consultancy rather than a scalable software platform to help enterprise clients build tools.
Biggest teaching moment ▶ 27:36 Explaining LLM evaluation through domain expertiseStanislas educates the host on why casual ChatGPT users mistake LLMs for search engines, showing how testing models on topics where you are a domain expert exposes hallucinations instantly.
Matt holds his own ▶ 46:17 Demonstrating deep knowledge of Paris AI research historyMatt demonstrates substantial background knowledge by detailing Yann LeCun's establishment of Meta's FAIR research lab in Paris and its role in cultivating French AI talent.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Stanislas Polu's Stanford Days and First Startup Journey | 2 | 3 | 0 | 0 | Matt asks open conversational questions about Stanislas's background at Stanford and early startup pivots. Stanislas comfortably recounts his early mistakes, such as choosing Oracle over Facebook and shifting between coupon apps and photo analytics. The dynamic is fully collaborative and conversational. | |
| The Stripe Acquisition Story and Early Growth Phase | 2 | 5 | 2 | 0 | Stanislas politely rejects Matt's framing of a bidding war between Pinterest and Stripe, clarifying the sequence of events. He recounts how Patrick Collison initially rejected them for being non-US until David Mazieres intervened. Matt listens adaptively as Stanislas corrects the narrative. | |
| Key Operating Culture Learnings from Stripe | 2 | 4 | 0 | 0 | Matt asks about early culture lessons from Stripe that apply to Dust. Stanislas explains how early Stripe operated with high talent density, no dedicated product managers, and open mailing lists. The interaction is an open, informative exchange. | |
| Joining OpenAI and the Dynamics of Frontier Research | 3 | 5 | 0 | 0 | Matt asks how OpenAI's research and engineering teams interact in practice. Stanislas provides detailed insight into compute allocation as an implicit alignment mechanism for researchers. The dynamic is respectful with Stanislas sharing insider operational details. | |
| An Insider-Outsider Perspective on OpenAI's Future and AI Scaling | 4 | 4 | 1 | 0 | Matt demonstrates topic awareness by referencing recent OpenAI valuation rumors ($90B) and Sam Altman AGI rumors. Stanislas carefully distinguishes his former insider perspective from his current outsider lens while explaining why AI scaling laws keep OpenAI's expected value high. The dialogue is balanced and analytical. | |
| The Founding Vision Behind Dust: Enterprise AI Product Packaging | 2 | 4 | 0 | 0 | Matt prompts Stanislas on why he left OpenAI to start Dust. Stanislas explains the gap between raw model power and enterprise product packaging, noting how company management are early adopters while internal staff follow traditional adoption curves. | |
| Overcoming Enterprise AI Traps with Constrained Tooling | 4 | 5 | 1 | 2 | Stanislas reframes Matt's assumption that ChatGPT turned everyone into power users, explaining how non-experts fall into traps when using unconstrained assistants. Matt pushes back by asking whether Dust is forced to act as a custom services business to help clients scope tools, which Stanislas addresses with a product-led builder strategy. | |
| RAG Architecture, Developer Capabilities, and Modular Assistants | 4 | 5 | 0 | 0 | Matt brings up technical concepts like RAG architecture and vector databases. Stanislas provides a technical deep-dive into semantic search limitations, structured data retrieval challenges, and model sensitivity to context noise. | |
| Multi-Model Agnosticism and Dust's Product Roadmap | 4 | 4 | 1 | 1 | Matt asks if automated prompt routing between models represents a standalone startup opportunity. Stanislas offers mild pushback, explaining that power users prefer raw frontier models over black-box routing layers for general team productivity tasks. | |
| The Boom of AI in France and Europe | 5 | 4 | 0 | 0 | Matt demonstrates detailed expertise on the European AI ecosystem, naming companies like Mistral and Synesthesia, as well as Yann LeCun's Meta FAIR lab in Paris. Stanislas expands on Paris talent density advantages over San Francisco while acknowledging the European capital funding gap. |
Statements from this episode (14)
Stanislas Polu's startup was Stripe's first European customer
“We knew them because we were the first European users.”
Polu: Stripe had no product managers when it reached 80 employees
“When we joined at People. There were very limited number of managers, very limited number of product manager. I think it was a forbidden word at the time. So no product manager that, that function was kind of covered by engineers and managers or engine manager…”
Polu: Compute allocation naturally aligns AI researchers with company goals
“There is a way to Orion's an organization, a research organizations, the way you allocate computes. Which means that as a researcher, it's often the case that you are free to work on whatever the things you want to work on. Right. But if the things you're work…”
Polu: OpenAI combined research and engineering early with fluid career ladders
“I think I saw the transition from research to research plus engineering being the ingredients into building the largest and best models. But even very early on OpenAI was very open on the fact that you could be one or the other Quite fluidly and changed betwee…”
Stanislas Polu: If transformative AI arrives in 5 to 10 years, OpenAI will likely build it
“But if it takes five years or 10 years, then that's probably where it's gonna happen.”
Polu: Onboarding enterprise users to AI requires constrained tools, not general assistants
“For a large amount of the people within companies, The best way to onboard them the technology is to not give them a very general assistant that can do anything and everything, but instead focus the assistant to some use case that feels more like a tool.”
Polu: Enterprise AI adoption will scale via product, not custom services
“I think all of that can probably scale through product and not necessarily handholding and kind of custom work with our clients eventually.”
Noisy retrieval causes even state-of-the-art LLMs to skip relevant context
“If what you're searching is too large, the answers, the chunk that you'll be finding will be a bit noisy, and the models, even the best ones today, they have a tendency to get a little bit disturbed by that noise, meaning that they might not hallucinate too mu…”
Polu: A single all-knowing enterprise AI assistant is currently far-fetched
“Context size and model quality makes it such that an assistant that knows it all within a company is, is, is, is still kind of a bit farfetched because it'll get confused if it sees too much information.”
Polu: AI model costs will fall, making heavy token usage profitable
“Because the cost will go down. So that's fine. Even if you're losing money today, you'll be winning some, I mean, earning some money tomorrow.”
Polu: Enterprise AI needs frontier models rather than complex query routing
“Most of the tasks are pretty general, right? Most of the tasks are pretty like a human would do. And so you just want the best models. And as it happens today, the best models are before enclosed. So that's what you want.”
Polu: Dust is near repeatable sales but has not reached PMF yet
“I think we are in the, we are on the precipice of repeated sales processes. So that could call PMF. I wouldn't quite qualify it as PMF yet.”