Key Moments
The Model-Agnostic AI Platform Betting That No Single Lab Will Win
Y CombinatorWant to know something specific about what's covered?
We've already dissected every moment. Ask and we will deliver (with timestamps).
Key Moments
Dust, a model-agnostic AI platform, competes with frontier labs by offering flexibility and integration, not by building its own models.
Key Insights
Dust's co-founder left OpenAI to focus on product building, despite lucrative stock options that were worth more than the entire company at one point.
The company aimed to apply LLMs to the workplace in early 2023, a niche that is now rapidly evolving.
Dust operates as a horizontal platform, contrasting with verticalized AI products, and is model-agnostic to provide flexibility to users.
The AI market's funding is heavily absorbed by large labs, making it challenging for smaller companies to raise capital.
Dust maintains reasonable valuations by focusing on product layer development and adopting a 'no GPU before PMF' mantra.
Margins for AI product builders are compressed due to token-based pricing from labs; Dust is transitioning to credit-based pricing to sustain profitability.
Transitioning from frontier research to product development
Stanislas Polu, co-founder of Dust, shared his journey from early-stage roles at Stripe and OpenAI to founding his own company. He explained that his departure from OpenAI, despite potentially significant financial gains (giving up stock options worth more than the company's valuation at one point), was driven by a desire to return to product building. Polu described research as an engaging but fleeting pursuit, where breakthroughs provide short-lived highs. He found himself wanting to build something more tangible and product-focused, a sentiment that guided his next venture.
The evolving vision for Dust and its market impact
Dust's initial motivation was to apply Large Language Models (LLMs) to the workplace, a move considered niche in early 2023. Polu noted that the definition of a company is a moving target in the rapidly evolving AI landscape, unlike the stable technological substrates of the past like JavaScript and PostgreSQL. He predicted that work will continue to evolve, with future generations viewing current work as significantly less demanding. While deeply convinced that AI would disrupt work, Polu admits he was wrong about the timeline for a plateau in technological advancement, underestimating the continuous pace of innovation. This continuous evolution means that even established players have decades of deployment and innovation ahead of them.
Embracing a horizontal, model-agnostic platform strategy
Dust intentionally adopted a horizontal platform strategy rather than a verticalized one, a choice that presents go-to-market challenges due to its broad applicability versus the focused appeal of vertical solutions. Polu likens Dust's approach to productivity suites like Notion or Google Drive but for human-agent interactions. This horizontal approach is complemented by being "model agnostic." Polu uses an analogy of energy providers: users should not be locked into a single AI model provider, especially if that provider has issues. This model agnosticism allows users to connect to different AI intelligences as needed, preventing dependency on a single, potentially problematic, supplier. This approach is seen as a key differentiator against the frontier labs that develop and offer their own specific models.
Navigating funding challenges amidst lab dominance
Polu acknowledged the significant impact large AI labs have on the funding market, drawing out investment capital. He noted that his own fundraising, particularly the Series B, was more challenging than earlier rounds. The perception is that investors might prefer to invest a billion dollars into a single large lab with the potential for massive returns, rather than distributing smaller amounts across multiple startups. This concentration of capital makes it harder for companies like Dust to secure funding. Polu expressed hope that successful labs going public could eventually open up funding avenues for other startups by creating market liquidity and reducing VC focus on frontier labs.
Strategic approach to valuation and funding
Dust deliberately pursued a strategy of raising at reasonable valuations, learning from friends who were 'burned' by raising too much in the 2021 boom. Polu emphasized a mantra: 'No GPU before PMF' (Product-Market Fit). They raised a modest $5 million seed round, focusing on product development rather than speculative training infrastructure, which was fashionable at the time. This approach allows them to build momentum and raise subsequent rounds when they are better positioned. While potentially perceived as cautious ('too French'), this strategy aims to avoid the 'coffin corner' scenario of raising too much capital without realizing its potential, which can lead to a forced downturn.
Building in France: A deliberate choice with its own challenges
Despite past experience building in San Francisco, Polu and his co-founder chose to build Dust in France, driven by a desire to contribute to their home country and a sense of national sovereignty. While acknowledging that building in the US would have been 'easier' from a company-building perspective, they believe it wasn't necessarily the 'right' thing for them. They note that when companies succeed, like Lovable or Spotify, their location becomes less relevant to funders. However, building in France introduces additional friction, a trade-off they accepted for a 'greater good,' suggesting this choice adds complexity but is a deliberate one for their long-term vision.
The future of verticalized AI products and defensibility
Polu discussed the trend of large labs entering new verticals, such as Anth
Mentioned in This Episode
●Software & Apps
●Companies
●Concepts
●People Referenced
AI Model Latency and Margin Comparison
Data extracted from this episode
| Model | Latency | Implied Margin |
|---|---|---|
| Cloudflare | Max acceptable latency for Cloudflare | Implied 70-80% margin for frontier models |
| GLM 5.2 | Roughly the same as Cloudflare | Implied 70-80% margin for frontier models (approx. 9x cheaper than frontier models with similar latency) |
Common Questions
Dust is a company founded by Stan, who previously worked at Stripe and OpenAI. Its mission is to apply Large Language Models (LLMs) to the workplace, focusing on creating horizontal platforms that integrate across various tools and workflows.
Topics
Mentioned in this video
Stan's previous employer where he spent five years.
Stan previously worked here for three years on large language models and maths.
A major AI lab, mentioned as a competitor to Dust and a significant player in the market.
Mentioned for its productivity suites, paralleled with Dust's approach to human-agent interactions.
A European AI lab mentioned as an example of successful company building in Europe.
Mentioned as a successful company that proves location is not a barrier for VC funding when things work well.
Mentioned as an example of Anthropic's expansion into new verticals, specifically the legal space.
Mentioned for its S1 filing, used as a reference point for understanding humongous margins in frontier models.
Mentioned for comparing the latency and cost of its models against GLM 5.2.
The company Stan co-founded, focusing on applying LLMs to the workplace.
Mentioned for its productivity suites, paralleled with Dust's approach to human-agent interactions.
An example of a horizontal platform company mentioned in the context of Dust's market.
An example of a horizontal platform company mentioned in the context of Dust's market.
Mentioned as a successful company that proves location is not a barrier for VC funding when things work well.
A model whose latency is compared to Cloudflare to illustrate profit margins in frontier models.
More from Y Combinator
View all 605 summaries
29 minWhy Ambitious Startup Ideas Are Actually Easier To Sell
75 minThe Key Thing Human Brains Have That AI Is Trying To Learn
33 minIndia Can Create The Largest AI Companies
41 minZynga Founder: Consumer Is Not Investible Right Now - Thats Why You Should Build It
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free