Eden Makes Fake Data That Looks Real So Your Sales Demos Stop Looking Empty

The Macro: The Demo Environment Problem Is Embarrassingly Universal

Here is something that happens at nearly every B2B software company. A sales engineer needs to give a product demo. The demo environment is either empty, broken, or full of test data that says things like “John Doe” and “Acme Corp” and “[email protected].” The prospect is supposed to imagine what the product looks like with their data in it. They cannot. The deal stalls.

This problem is so common that it has become invisible. Sales engineering teams spend an absurd amount of time maintaining demo environments. Some companies have dedicated people whose entire job is keeping demo instances populated with believable data. Others use production data, which creates obvious compliance and privacy nightmares. Most just wing it and hope the prospect has enough imagination to see past the empty dashboards.

The synthetic data market has been growing fast, but most of the investment has gone toward two use cases: training machine learning models and privacy-compliant analytics. Companies like Mostly AI, Gretel, and Tonic are focused on generating statistically accurate datasets that preserve the properties of real data without exposing personal information. That is valuable work, but it is solving a different problem than what sales teams face.

Sales demos do not need statistically accurate data. They need data that looks convincing to a human sitting in a conference room. The users need realistic names, plausible transaction histories, believable usage patterns, and enough variety that the product feels alive. This is more of a simulation problem than a privacy problem, and nobody has built a dedicated tool for it.

The adjacent problem is AI agent evaluation. If you are building agents that interact with customer data, you need test scenarios that cover edge cases, adversarial inputs, and multi-step workflows. Generating those by hand is slow and incomplete. You always miss the weird case that breaks things in production.

The Micro: Two Datadog Engineers Who Lived This Problem

Eden generates production-quality synthetic data for three use cases: sales demos, AI agent testing, and model training. You describe the domain and the data schema, and it generates realistic users, transactions, conversations, and whatever else your product needs to look populated and functional.

Alex Talamonti and Jason He founded the company. Both worked at Datadog, which is relevant because Datadog is exactly the kind of product that needs convincing demo data. You cannot demo an observability platform with empty charts. Alex focused on data visualization and analysis there. Jason is a previous founder with an exit. They are a two-person team in New York, part of Y Combinator’s Summer 2025 batch.

The product handles demo-specific generation with schema awareness. You tell it “I need data for a sales call platform” and it generates realistic call logs, deal stages, contact records, and activity timelines. Not random noise. Contextually appropriate data that looks like it came from a real company’s database.

For AI agent evaluation, Eden generates adversarial test cases. Ambiguous inputs, edge cases, complex multi-step scenarios. The kind of stuff that is tedious to write by hand and critical to get right before deploying an agent to production.

The training data angle covers instruction pairs, chain-of-thought traces, and RLHF preference data. This puts Eden in competition with Scale AI and Surge AI on the labeling side, though at a different price point and for a different customer profile. Scale is enterprise. Eden is targeting teams that need hundreds to thousands of examples, not millions.

The go-to-market is currently demo-driven. No public pricing, book-a-call model. The site is live and built on Next.js. The value proposition is clear: “Synthetic data that feels real. Generated in seconds, not sprints.”

The Verdict

I think Eden is solving a problem that every B2B SaaS company has and almost none of them have solved well. The demo environment is one of those unglamorous bottlenecks that costs more deals than anyone wants to admit. If Eden can generate convincing, domain-specific data in seconds rather than days, sales engineering teams will pay for it without hesitation.

The risk is scope. Three use cases (demos, agent testing, training data) is a lot for a two-person team. Each has different quality requirements and different buyers. Demo data needs to look good to humans. Agent test data needs to be adversarial and comprehensive. Training data needs to be accurate and diverse. I would want to see them nail one of these before spreading across all three.

In thirty days, I want to know how many sales engineering teams are using it in production and whether the generated data actually holds up during live demos. In sixty days, I want to see if the agent testing use case has traction or if it is a feature looking for a market. In ninety days, the question is whether Eden can build enough domain templates that new customers get value immediately, or whether every deployment requires custom configuration. Speed to value will determine whether this is a tool people adopt or a tool people evaluate and forget.