$18.5M from CRV and Amplify to sell judgment. Design is the beachhead, not the business — which is the difference between this and everyone else in the lane.
Latest (Sep 2026): Came out of stealth on 16 Jun 2026 with an $18.5M seed co-led by CRV and Amplify Partners. Latitud wrote the first cheque and Gokul Rajaram publicly backed it. The founding team is about 20 people from Exa, Palantir and Mercado Libre, including a VP of Design with 15 years' experience. Its job ads claim 'most frontier labs' as existing customers, which is unverified. On 16 Aug 2026 it published Requests for Research and launched the Prototype research fellowship. As of Sep 2026 it is hiring AI engineers (RL, backend, fullstack) at $175–275K plus equity, a Growth Lead and a remote Community Operations Manager for its 'Taste Makers' designer network.
Taste Labs emerged from stealth with an $18.5M seed co-led by CRV and Amplify Partners, founded by Thais Castello Branco, formerly growth lead at Exa, with a founding team of about 20 people from Exa, Palantir and Mercado Libre plus a design VP with fifteen years in the discipline (The Source Code).
The one structural fact that separates it from Contra Labs and Design Arena is in the positioning, not the product: design is explicitly the initial vertical, a beachhead, not an end state. Everyone else in Design and UI/UX is building a design company. Taste Labs is building a judgment company that starts in design because design is where the deficiency is most visible.
Two things to two different buyers. To frontier labs: preference datasets, reasoning data, rubrics and evaluation environments. To application companies: verification software (The Source Code).
That is a materially better shape than the pure data-foundry model. The data sells into the lab budget, which is deep and lumpy; the software sells into the application layer, which is shallow but recurring and does not evaporate when a lab finishes a training run. It is the answer to the objection that hangs over every specialist here — that design data is a project rather than a budget line.
Amplify's investment memo carries the strongest demand evidence in the design vertical, and it is not a survey or a benchmark. Runway, Figma and Adobe found that optimising models against general user preference degraded quality as judged by experts. They needed feedback from tastemakers, not from average users (Amplify Partners).
That is a specific, named, technical failure at three real companies, and it is a far better sales argument than any leaderboard. The memo's line — "when judgment is scarce, slop fills the void" — is the category thesis in six words.
Note what it also implies. The buyers named are tool vendors, not frontier labs. As with Contra Labs' corroborated partner list, the corroborated demand in design comes from the application layer. See How much money is actually in the buyer pool before assuming the nine-figure lab budget applies.
Expert sourcing runs on a trust-propagation model: seed with a small set of vetted tastemakers, then let them nominate others (The Source Code).
This is the cheapest known answer to Which side you build first for a company with no captive community. Contra Labs solved the same problem by inheriting 1.5M marketplace profiles from a parent that spent six years and $45M assembling them. Taste Labs cannot inherit, so it borrows the vetting instead of performing it — each nomination is a credential issued by someone whose taste has already been accepted.
The trade-off is honest: referral graphs are fast, cheap and self-limiting. They saturate inside a professional clique, and the network you get looks like the seed you chose. Contra's marketplace is broader and shallower; this is narrower and deeper. Neither is disclosed as a number, so nobody can yet say which produces better inter-rater reliability.
| Taste Labs | Contra Labs | Design Arena | |
|---|---|---|---|
| Capital | $18.5M seed | ~$45.2M at the parent | $7.9M seed |
| Supply | Trust propagation from vetted seeds | Inherited marketplace, 1.5M/1.7M claimed | 5.3M consumer users |
| Product | Preference data + rubrics + verification software | Preference pairs, trajectories, benchmark | Crowd preference arena |
| Scope | Design as beachhead | Design, explicitly | Design, explicitly |
| Revenue | Not disclosed | Not disclosed | $60M ARR claimed |
Three companies, three different theories of where the scarce input lives: in a curated clique, in an owned marketplace, in a consumer crowd. All three are selling into the same deficiency.
No revenue, no ARR, no customer names, no valuation, no headquarters, no founding date beyond "emerged from stealth". The trust-propagation network has no disclosed size. Nothing is published about pricing or contract shape. The Runway/Figma/Adobe finding comes from the lead investor's own announcement post — it is the best demand evidence in the vertical and it is also marketing written by a party with a position.
What to learn from it: a beachhead framing raises more money than a vertical framing, because it prices the second market for free — and the referral graph is the only supply mechanism here that works without owning a community first.
The bet is that one domain buys cheaper experts and faster belief, and that both advantages expire the moment you have a reference customer. What would have to be true, what the evidence supports, and the trade-off that decides which niche.
5.3M users, $60M ARR, under a year old — the consumer-arena route into the same data. Nobody has said whether that number is gross or net.
The thesis that started this whole enquiry, proven by three funded companies that between them already hold the network, the rubric science and the preference collection.
A rich pool with excellent free channels, sitting behind a platform holder who has already banned the thing you would need to do, in an occupation BLS projects at 0% growth.
Miju Labs
