A San Francisco startup pitching itself as the solution to artificial intelligence's "judgment problem" has raised $18.5 million in seed funding. Taste Labs announced the round on June 16, with CRV and Amplify Partners serving as co-leads and participation from Latitud, BoxGroup, Banter Capital, and several prominent angel investors including Gokul Rajaram and Logan Kilpatrick.
The company sells to two distinct customer bases. Foundation model labs buy post-training datasets and reinforcement learning environments from Taste. Application developers, meanwhile, use its Brand API to ensure their generative AI agents adhere to design guidelines and company style rules.
Taste declined to share its valuation. Industry tracker Dealroom pegged it at $74 million, though that figure could not be independently confirmed.
Selling Taste to Machines
Founder Thais Castello Branco, who previously led growth for the Websets product at search startup Exa, describes her company's mission in terms that blur the technical and aesthetic. "We're affecting this via post-training data and RL environments," she wrote in a manifesto published at the company's launch. The goal, she says, is raising quality baselines in subjective tasks before optimizing for individual preferences.
Sarah Catanzaro, a partner at Amplify Partners, put it more bluntly in her firm's investment announcement. "Generative AI has made creation cheap," she said. "What it has not made cheap is judgment."
That gap has created commercial opportunity. Taste's Brand API, which went live publicly in September, offers three functions. The Extract endpoint pulls brand assets. Search finds design references. Verify checks outputs against style rules. Developers access the API through standard HTTP calls or via Model Context Protocol servers compatible with Claude, Cursor, and similar tools.
Pricing starts at $99 monthly for 600 credits, climbing to $749 for 6,000 credits. Enterprise tiers exist for larger customers, though the company has not published those rates.
The Contractor Network
Behind the API sits what Taste calls its TasteMakers network, which the company says includes over 1,000 designers spread across 30 countries. These contractors generate the subjective judgments that become training data. How Taste ensures consistency across that distributed workforce remains unclear, and the company did not respond to questions about quality control mechanisms.
In job postings, Taste claims "most frontier labs are already customers," though it has yet to name any foundation model providers publicly. That omission may be contractual rather than strategic; enterprise AI deals typically include strict nondisclosure terms.
One customer willing to go on record is General Intelligence Company, the New York outfit behind a product called Cofounder. CEO Andrew Pignanelli said in an October case study that his team used the Brand API to import design guidelines. "This was a harder, more obscure problem than we thought," he noted. "It was just fire and forget. It just worked great."
Whether that ease of integration holds at scale is another matter. Early-stage case studies tend to highlight best-case scenarios.
Competitive Context
The market for specialized AI training data has grown crowded. Scale AI and Surge AI represent larger, better-capitalized competitors in adjacent segments, and both have raised substantial sums over the past two years. What differentiates Taste, according to its backers, is focus on subjective domains where programmatic evaluation fails.
CRV partners Veronica Orellana, Reid Christian, and Mia Krishnamurthy wrote in their announcement that they view the startup as infrastructure for "taste" itself, starting with design but eventually expanding to other creative fields. That vision assumes taste can be systematized, a proposition that might make traditional designers uneasy.
Latitud, a Latin American venture fund, said it wrote the first check into Taste months before the seed round closed. CRV's portfolio materials indicate the firm partnered with the company in April, roughly six months after its founding. Taste reported 18 employees at the time of the funding announcement. LinkedIn data suggested headcount in the 11-50 range as of October, according to RL List, an industry tracker.
The startup has also moved into research territory. It launched a fellowship program called Prototype in August and published a requests-for-research agenda covering machine creativity, design evaluation, and agent behavior. Taste presented work at the AI Engineer Conference in September, though details of that presentation were not made public.
What Comes Next
Whether Taste can maintain dual revenue streams as it scales is an open question. Selling to foundation labs and application developers requires different sales motions, technical capabilities, and customer support structures. The company will need to prove it can serve both markets without diluting focus.
For now, investors seem willing to bet on Castello Branco's thesis that AI's next bottleneck is not generation but discernment. Whether that bottleneck proves wide enough to support a venture-scale business remains to be seen.
