Mecka AI nears $500M. Recorded work is now inventory.

A company that pays people to film themselves making coffee is reportedly raising at a half-billion-dollar valuation. TechCrunch reports that Mecka AI is nearing a Sequoia-led round valuing it around $500 million, three months after a $60 million Series A. Terms are not final and Mecka did not comment, so treat the valuation as reported rather than settled. The direction is the part that matters: recordings of ordinary work are being repriced as training data, and most businesses are giving theirs away.

What actually happened

Mecka collects human motion data to train humanoid robots. Per Fortune's coverage of the Series A, the company pays people to record themselves doing everyday tasks — making coffee, fixing cars — using body sensors and smartphones. That $60 million round in June 2026 was led by Framework Ventures, with Menlo Ventures, SV Angel, and Kindred Ventures participating. Co-founder Josh Gao projected an annual run rate of $100 million by the end of 2026.

The model is the same one Scale AI, Mercor, and Surge built for language models, pointed at the physical world. Text was scraped; physical demonstration has to be produced. That scarcity is what a Sequoia term sheet three months after a Series A is pricing.

Why your process data matters for your business

You are not selling robot training data. But the same scarcity logic applies one level up, and it is the reason most small-business automation projects stall.

When you try to automate a workflow — quoting, intake, dispatch, returns triage — the blocker is almost never the model. It is that nobody wrote down how the work is actually done. The real process lives in one person's head, includes four exceptions that never got documented, and differs from the SOP in ways that only surface when the agent gets it wrong in front of a customer. Demonstration data is the expensive input. Mecka's valuation is a market putting a number on that.

So capture it while you have the people who know it. Recorded screen sessions of a senior person doing the task end to end. The decision points, stated out loud. The exceptions, with the reason. Transcripts of the calls where the judgment call happens. That corpus is what turns an agent from a script with a chat interface into something that handles your edge cases, and it is worth more than whatever tool you were about to buy.

Then keep it. Read the terms of every AI vendor you hand it to. If your support transcripts, call recordings, or annotated workflows become training data for a model your vendor sells to your competitor, you funded their product. Ask two questions in writing: do you train on our data, and can we export the corpus and our annotations in a usable format if we leave. Vague answers are answers.

Sources: TechCrunch, Fortune.