Ooak Data has quietly built a business most enterprises didn't know existed: converting their internal Slack messages, Gmail threads, and project management archives into testing grounds for artificial intelligence. The Y Combinator-backed startup announced it has already paid out $1 million to companies willing to hand over their data, according to cofounder Grégoire Lamy in a July 2024 post.
The pitch sounds counterintuitive in an era of data breaches and privacy scandals. But AI developers building autonomous agents have hit a wall. Systems trained on carefully crafted synthetic scenarios collapse when they encounter the messy reality of how work actually happens across multiple software tools. Ooak Data is betting that friction represents a market opportunity.
"Agents fail in production not because the models are bad, but because they've never trained on how work actually happens," CTO Thomas Aubry wrote in a recent LinkedIn post. That observation has shaped the company's entire strategy: source real enterprise workflows, strip them of identifying information, then sell access to AI labs that need authentic training environments.
The technical challenge involves more than just redacting names. Ooak Data's system ingests material from 18 different workplace platforms, including Gmail, Slack, Notion, Jira, Google Drive, SharePoint, Teams, GitHub, GitLab, Linear, Confluence, Figma, Miro, Salesforce, HubSpot, Trello, and Airtable. An automated pipeline removes identifying details while preserving the underlying structure of complex, multi-step processes that span several tools.
Perhaps more striking than the technology is the economics. The startup advertises payouts reaching $300,000 for particularly valuable datasets, according to its partner page. Companies must sign nondisclosure agreements before transferring data and undergo what Ooak Data describes as a "deep audit." European partners work under GDPR frameworks according to the company's partner page and privacy policy, with regional processing designed to keep EU data within EU borders.
What Ooak Data charges on the other side of that transaction remains undisclosed. The company has not revealed pricing for the AI labs and enterprises that license access to these anonymized training environments, nor has it publicly identified any customers. The startup describes its target market on its homepage broadly: frontier AI labs, enterprise AI teams, and venture-backed startups building autonomous systems.
The competitive landscape has emerged quickly. HUD offers reinforcement learning environments at ten cents per environment hour, according to its website. Foundry RL, currently in private beta according to its website, provides what it calls pixel-perfect reproducible browser environments for AI web agents. Centific launched a service it branded RL Environments-as-a-Service earlier this year, emphasizing branching, multi-step tasks meant to mirror real business conditions.
Y Combinator's latest batch included multiple startups tackling similar infrastructure challenges, among them CoArena and Olam Labs alongside Ooak Data. Industry momentum appears broader still; OpenAI recently updated its Agents SDK to accommodate third-party sandbox providers, a signal that the ecosystem around agent testing is maturing.
Pierre-Louis Vouteau, Grégoire Lamy, and Thomas Aubry founded Ooak Data in France in September 2024. The company maintains dual entities: Ooak Data SAS registered in Paris and a Delaware corporation with a Wilmington address. Y Combinator lists the team as operating from San Francisco with five people, though LinkedIn's self-reported range suggests between 11 and 50 employees. Those platform discrepancies are common and often reflect lag time in updates rather than intentional obfuscation.
Before Y Combinator, the startup participated in Station F's Founders Program. Funding details beyond the accelerator's backing have not been disclosed. Current hiring posts list roles in data engineering, software development, and marketing with equity grants ranging from 0.10 percent to 1.00 percent and salaries spanning €50,000 to $100,000.
The company has also contributed open-source tools to the broader research community. APEX Explorer, available at apex-explorer.ooakdata.com, provides an interface for browsing 480 tasks, traces, and prompts from Mercor's APEX-Agents dataset. Ooak Data published research essays earlier this year arguing that static benchmarks no longer capture how agents perform once deployed in production settings.
Vouteau framed the value proposition succinctly in a July 2024 LinkedIn post, describing the service as one that "turns real company workflows into environments where AI agents learn to work." The alternative, he suggested, is synthetic data that fails to reflect operational complexity.
The $1 million payout milestone Lamy announced suggests Ooak Data has begun securing data partnerships at meaningful scale, though how many companies that figure represents and what average payouts look like remains unclear. The company has declined to share customer names or per-dataset pricing structures.
Whether enterprises will continue feeding their internal processes into what amounts to a marketplace for corporate behavior patterns is an open question. Privacy concerns, competitive intelligence risks, and the simple strangeness of monetizing everyday work routines may limit supply. Then again, $300,000 is real money for companies sitting on data they previously considered inert.
