Ooak Data

An AI data-infrastructure startup that turns internal company data into anonymized training/evaluation environments for AI agents.

Website: https://ooakdata.com/

Cover Block

From the public record
Name Ooak Data
Tagline An AI data-infrastructure startup that turns internal company data into anonymized training/evaluation environments for AI agents.
Headquarters Wilmington, United States
Founded 2024
Stage Seed
Business Model SaaS
Industry Deeptech
Technology AI / Machine Learning
Geography Global / Remote-First
Growth Profile Venture Scale
Founding Team Co-Founders (3+)
Funding Label Seed (total disclosed ~$100,000)

Links

From the public record

The Short Version

From the public record Ooak Data is building data infrastructure that transforms real enterprise workflows into training environments for AI agents, a critical bottleneck for moving agents from demos to production [Ooak Data]. Founded in 2024, the company is part of Y Combinator's Summer 2026 batch and is positioned to address a growing need for realistic, privacy-safe evaluation data [Y Combinator]. The core product connects to tools like Slack, Gmail, and Jira, anonymizes the data into a structurally identical "digital twin," and provides a platform for labs to train and test agents on authentic tasks [Ooak Data].

The founding team, comprising Pierre-Louis Vouteau, Thomas Aubry, and Grégoire Lamy, brings operational experience from inside the complex company systems they aim to replicate [Ooak Data, Ooak Data]. Initial backing includes Y Combinator and Vela Partners, with a disclosed seed investment of $100,000, though the full capitalization remains private [Y Combinator]. The business model is structured as a SaaS platform, with a dual corporate entity in Delaware and Paris suggesting a deliberate strategy for U.S. and EU data compliance and market access [Ooak Data].

Over the next 12-18 months, the key milestones to watch are the formal announcement of a priced funding round, the disclosure of initial enterprise or AI lab customers to validate the product's security and utility, and the expansion of the team from its current five-person base [Y Combinator]. The company's ability to secure high-value data partnerships will be the primary determinant of its trajectory. Single-source, plausible -- Core product claims are from primary sources, but funding details and team size are from limited secondary corroboration.

Taxonomy Snapshot

Axis Classification
Stage Seed
Business Model SaaS
Technology Type AI / Machine Learning
Geography Global / Remote-First
Founding Team Co-Founders (3+)

The Company in Brief

From the public record

Ooak Data was founded in 2024 as an applied AI research lab, emerging publicly in August 2026 as a participant in Y Combinator's Summer 2026 batch [Y Combinator]. The company's founding premise is direct: its three co-founders, Pierre-Louis Vouteau, Thomas Aubry, and Grégoire Lamy, spent their careers inside the complex, often messy internal systems of companies and now aim to replicate that environment for AI training [Ooak Data]. They operate with a dual-entity structure, incorporating as Ooak Data, Inc. in Delaware for global operations and Ooak Data SAS in Paris to handle European data processing and compliance [Ooak Data].

While the company lists a Wilmington, Delaware address for its corporate registration, its operational footprint appears distributed. The Y Combinator profile notes a team of five employees based in San Francisco, and job listings point to active hiring in Paris, suggesting a remote-first, transatlantic model [Y Combinator][Ooak Data]. The early milestones are characteristic of a Y Combinator-backed venture: program participation, a public launch profile, and the opening of multiple technical and go-to-market roles within months of founding.

Single-source, plausible -- Core founding and entity details are confirmed via primary company sources, but headcount and location specifics rely on a single secondary source.

What They Have Built

Mixed sourcing The product is a data-infrastructure platform designed to close the gap between AI agent demos and real-world deployment. It connects directly to a company's internal tools,including Gmail, Slack, Notion, Jira, Drive, and SharePoint,and creates an anonymized, structurally identical replica of the workflow environment [Ooak Data]. This 'digital twin' serves as a privacy-safe training and evaluation sandbox where AI agents can be tested on tasks that reflect actual business operations, from navigating Slack threads to managing broken exports and approval chains [Ooak Data].

The company's public-facing demonstration, APEX Explorer, showcases the environment's capabilities. It presents a series of tasks within simulated worlds, such as a legal or production domain, where agents must execute shell commands in a sandbox, create and edit Excel spreadsheets, and interact with a filesystem [Ooak Data]. This suggests a focus on multimodal data and tool-use evaluation. The underlying technology stack is not detailed, but open roles for a Lead Data Engineer and Technical Ops Manager point to a need for robust data pipeline engineering and systems management (inferred from job postings) [Y Combinator].

Ooak Data positions its output as both datasets and evaluation environments for frontier AI labs. The research division publishes on evaluation methodology and dataset design, indicating the product is intended to generate not just raw data but also the frameworks to measure agent performance within these synthesized environments [Ooak Data].

Single-source, plausible -- Product claims are sourced from the company's own website and demo, with some technical inferences drawn from public job listings.

Market Size and Demand

From the public record

The market for AI agent training data is emerging from a fundamental mismatch: models that ace benchmarks fail at the unstructured, multi-step workflows that define actual business operations.

Quantitative market sizing for AI agent training environments specifically is not yet established in public third-party reports. Analysts have, however, begun sizing adjacent categories. The market for AI training data overall was valued at approximately $2.5 billion in 2023, with projections for high growth driven by demand for high-quality, domain-specific datasets [Grand View Research, 2024]. More directly analogous, the market for AI in business process automation, which agentic systems aim to power, is forecast to reach $19.6 billion by 2027 [MarketsandMarkets, 2023]. These figures suggest a substantial underlying economic activity that Ooak Data's proposition seeks to serve, though the specific wedge of anonymized, real-world workflow data for agent training remains unquantified.

Demand is driven by two converging trends. First, frontier AI labs and enterprise R&D teams have hit a wall with synthetic or static benchmark data, which lacks the nuance, tool integrations, and human-in-the-loop complexity of real work. Agents fail in production because they haven't trained on how work actually happens, including Slack threads, broken exports, and redundant tools [Ooak Data]. Second, enterprises are sitting on vast troves of internal workflow data but lack a secure, compliant mechanism to use it for AI development, either internally or as a monetizable asset. The company's dual-entity structure, with a Delaware corporation and a Paris SAS, is a direct response to the regulatory tailwind of stringent data sovereignty laws in the EU and elsewhere, positioning it to handle data where it originates [Ooak Data].

Key adjacent and substitute markets include providers of synthetic data generation and traditional business process outsourcing. Synthetic data vendors address the volume and privacy challenge but cannot replicate the emergent complexity of authentic human workflows. Business process outsourcing represents the incumbent, non-AI solution for workflow automation, against which AI agents must prove superior efficiency and cost. The regulatory environment is a primary macro force. Data protection regulations like GDPR in Europe and evolving AI governance frameworks globally create both a barrier and a moat for any solution involving sensitive corporate data; compliance is not a feature but a foundational requirement.

AI Training Data (Overall) 2023 | 2.5 | $B
Business Process Automation AI 2027 | 19.6 | $B

The available sizing data, while not specific to Ooak Data's niche, indicates the scale of the adjacent markets it intends to penetrate. The nearly eight-fold projected growth in business process automation AI underscores the significant budget allocation and strategic priority enterprises are placing on automating complex workflows, which is the core problem Ooak Data's data infrastructure aims to solve.

Single-source, plausible -- Market sizing is drawn from analogous, published third-party reports. The specific market for anonymized workflow training data is not yet independently sized.

Who Else Is Fighting for This

Mixed sourcing Ooak Data is positioned as a specialist in a nascent, high-friction niche: sourcing and anonymizing real enterprise workflow data to create training environments for AI agents, a segment that sits between data infrastructure and AI evaluation platforms.

The primary alternatives for an AI lab seeking training data are not single-platform competitors but a collection of point solutions and internal efforts.

  • Incumbent data platforms and marketplaces. Established players like Scale AI and Labelbox focus on data annotation and management for supervised learning, not on constructing interactive, anonymized replicas of corporate tool ecosystems. Their edge is in scale, trust, and existing enterprise contracts, but they do not offer the specific product of a connected, privacy-safe digital twin. Data marketplaces (e.g., Snowflake Marketplace, AWS Data Exchange) aggregate third-party datasets, which are typically static and lack the interactive, tool-connected environment Ooak Data proposes.
  • Agent evaluation and benchmarking startups. A growing cohort of companies, such as Weights & Biases (through its LLM evaluation tools) and more recent entrants like Langfuse or Phoenix, provide frameworks for evaluating AI agent performance. Their focus is on the evaluation layer itself, often assuming the test environment is already built. Ooak Data’s wedge is supplying the foundational, realistic environment those evaluators need, positioning it as a potential data supplier to this layer rather than a direct rival.
  • Internal development. The most significant competitive threat is in-house teams at large AI labs or enterprises building custom data pipelines. This path offers maximum control and data privacy but requires substantial engineering investment and domain expertise in both data engineering and anonymization. Ooak Data’s value proposition is to productize this complex, cross-disciplinary build.

Ooak Data’s defensible edge today appears to be its early focus on the full-stack problem of data sourcing, anonymization, and environment creation. The company’s public materials emphasize a research-driven approach to dataset design and evaluation methodology [Ooak Data], which could translate into proprietary techniques for structuring synthetic yet realistic workflow data. Furthermore, its dual corporate structure,a Delaware corporation and a Paris SAS,signals a deliberate strategy to navigate EU and US data regulations [Ooak Data], a compliance moat that could be significant for handling sensitive enterprise data. However, this edge is perishable; it depends on securing initial enterprise data partnerships to build a unique dataset before larger, well-capitalized infrastructure players decide to expand into this specific use case.

The company is most exposed on two fronts. First, on distribution and sales: securing the initial enterprise data partnerships requires a sophisticated go-to-market motion that trusts a young startup with a company’s most sensitive operational data. Second, on technological scope: if leading AI labs conclude that high-quality synthetic data generation (from companies like Mostly AI or Hazy) is sufficient for agent training, the need for a platform that physically connects to and anonymizes real tools could be circumvented. A competitor with superior synthetic data generation for complex workflows could emerge as a substitute.

The most plausible 18-month scenario hinges on Ooak Data’s ability to convert its Y Combinator network into a handful of flagship enterprise data partners and lab customers. If they succeed, they become the de facto data infrastructure partner for a small but influential set of frontier labs, creating a network effect where their environment becomes the standard testbed. The winner in this case would be Ooak Data, carving out a defensible niche. If they fail to secure those critical early partnerships, the loser would be Ooak Data’s current business model, as the market may conclude the integration and trust hurdles are too high, leaving the space to be addressed piecemeal by larger data platforms expanding their offerings or by labs building in-house solutions.

Single-source, plausible -- Competitive analysis is inferred from adjacent market segments and company positioning; no direct competitors are named in public sources.

Opportunity

From the public record The prize for Ooak Data is to become the primary supplier of the real-world, permissioned data that determines whether AI agents can function in enterprise environments, a role analogous to a foundational infrastructure provider for a new computing paradigm.

The headline opportunity is for Ooak Data to become the default data-infrastructure layer for AI agent development, a category-defining platform that sits between enterprise data and frontier AI labs. The company's foundational premise, that agents fail in production because they haven't trained on how work actually happens, addresses a critical and widely acknowledged bottleneck [Ooak Data]. By focusing on anonymizing and structuring data from core workflow tools like Slack, Gmail, and Jira, they are not just selling a dataset but a repeatable process for generating high-fidelity training environments. This outcome is reachable because the need is already articulated by the AI research community, and Ooak Data's early public work, such as the APEX Explorer platform with its detailed simulation worlds, demonstrates a technical approach beyond a conceptual slide [Ooak Data]. Their dual corporate structure for US and EU data compliance further signals a foundational, rather than tactical, approach to securing sensitive enterprise data partnerships [Ooak Data].

Growth is not a single path but a branching set of scenarios, each with a distinct catalyst.

Scenario What happens Catalyst Why it's plausible
The Evaluation Standard Ooak Data's environments become the industry benchmark for pre-production agent testing, adopted by major labs and enterprises. A public partnership with a leading AI lab (e.g., OpenAI, Anthropic) to co-develop and endorse an evaluation suite. The company publicly positions its research on the gap between benchmarks and real-world capability, framing itself as a solution to a known problem [Ooak Data].
The Data Monetization Engine Enterprises adopt Ooak Data primarily as a secure channel to license their anonymized workflow data to AI developers, creating a new revenue stream. Securing a flagship partnership with a large enterprise (e.g., a Fortune 500 company) willing to be a named case study. The company's messaging explicitly targets enterprises seeking to "safely monetize" internal data, indicating this is a designed go-to-market path [Y Combinator].
The Vertical Specialist The company achieves dominance in a specific high-stakes vertical (e.g., legal, finance) by building deeply specialized, compliant data environments. The launch of a vertically tailored product suite, backed by domain-specific research and compliance certifications. Their APEX Explorer demo worlds already showcase complex, narrative-driven tasks in domains like law, suggesting an initial focus on depth over breadth [Ooak Data].

Compounding for Ooak Data would manifest as a data network effect and an increasing returns-to-scale product. Each new enterprise data partnership expands the diversity and realism of the training environments, making the platform more valuable for AI labs seeking robust evaluation. In turn, adoption by more labs creates stronger demand signals for enterprises to contribute data, creating a two-sided marketplace. Early evidence of this flywheel is not yet public in the form of named partners, but the company's architecture is built for it. Their platform is designed to connect to multiple company tools and anonymize data at scale, a process that likely becomes more efficient and defensible with each new data schema ingested and each new privacy-preserving technique developed [Ooak Data].

Quantifying the size of the win requires looking at comparable infrastructure plays in adjacent data markets. Snowflake's initial value proposition was not just data storage but enabling secure, governed data sharing and monetization, a paradigm that reached a market cap exceeding $50 billion at its peak. While Ooak Data operates in a nascent, more specialized niche, a successful execution of the "Data Monetization Engine" scenario could position it as the Snowflake for AI training data. If the company captured even a single-digit percentage of the spending by AI labs on data acquisition and synthetic environment building,a market estimated to be in the billions annually,it could support a multi-billion dollar valuation. This is a scenario-based outcome, not a forecast, but it illustrates the scale of the infrastructure opportunity they are attempting to capture.

Single-source, plausible -- The opportunity framing relies on the company's stated technical approach and market positioning, which are well-documented in primary sources. The growth scenarios and comparable analysis are logical extrapolations from this foundation, but lack third-party validation or disclosed commercial traction.

Sources

From the public record

  1. [Ooak Data] Ooak Data , Applied AI Research Lab | https://ooakdata.com/

  2. [Y Combinator] Ooak Data: We turn company data into training data | https://www.ycombinator.com/companies/ooak-data

  3. [Grand View Research, 2024] AI Training Data Market Report | https://www.grandviewresearch.com/industry-analysis/ai-training-data-market-report

  4. [MarketsandMarkets, 2023] AI in Business Process Automation Market Report | https://www.marketsandmarkets.com/Market-Reports/ai-business-process-automation-market-247658558.html

Articles about Ooak Data

Related companies

Topics

View on Startuply.vc