Miju Labs

← All buyers · Figma

David Kossnick

Head of AI Products (also styled Head of AI / Head of Product, AI) · Figma

high confidenceTier A · Talk to firstDesign & UI productsUI · image

Kossnick runs all AI products at Figma (Figma Make, First Draft, AI search, Weave) and personally designed the evaluation process behind Figma Make. That process contains the single most explicit public statement by a design company that it pays human contractors to run nightly 'taste and judgement' evals against internal brand guidelines. He is the clearest named buyer of design-taste data in the whole index.

Background

Product career at Google and then YouTube, after which he co-founded and was CEO of LatentSpace and later ran AI at Coda before joining Figma to lead AI products. That path — consumer product management, then a generative-AI startup, then AI inside a design tool — explains his habit of treating model quality as a product-research problem rather than a benchmark problem: he ran company-wide 'bake offs', 1,000-example FigJam labelling sessions with PMs and designers, and concentric rings of testers from 30 internal AI staff out to external alpha customers.

He was named to Fast Company's AI 20 for 2025. In interviews he is consistently the 'humans stay the pilot' voice at Figma: craft, empathy and taste are what remain scarce as generation gets cheap, and AI collaboration should look like brainstorming with a teammate rather than a terminal prompt.

What they run now

Owns Figma's AI product surface — Figma Make (prompt-to-app), First Draft, AI-assisted code layers, the Dev Mode MCP server, and now the Weave canvas from the Weavy acquisition — plus the evaluation machinery that gates them. Figma Make shipped against two scored metrics: a 1-4 design score (does the output match the mockup or the prompt visually) and a 1-4 functionality score. Four eval types run underneath: deterministic pass/fail scripts, 'taste and judgement' evals in which human contractors grade outputs nightly against Figma's internal brand-quality guidelines across competing system prompts and model choices, AI-as-judge prompts calibrated on human examples, and A/B usage analytics on production cohorts. Figma wrote its own quality-rater guidelines for those contractors, explicitly modelled on Google's search quality evaluator guidelines.

What they buy

Confirmed: contractor-run human evaluation of generated UI and app outputs, scored on design quality and functionality against written brand-quality guidelines, run nightly across model and prompt variants. That is exactly a rubric-plus-rater product. Strongly implied but not confirmed: preference pairs over competing generations for reward modelling or model selection, calibration sets for AI-as-judge, and designer-authored critique of generated UI. A July 2026 Figma job ('Product Designer, AI Models', SF/NY/remote, $169k-$303k) makes the vendor relationship explicit: the hire is an 'arbiter of craft and taste' who oversees the quality of design evaluations and gives guidance to external partners while managing contractors. No vendor is named publicly.

Career

  1. presentHead of AI Products, Figma — Titled 'Head of Product, AI' in May 2025 and 'Head of AI' in Dec 2025 coverage.
  2. —Head of AI, Coda
  3. —Co-founder & CEO, LatentSpace
  4. —Product Manager, YouTube
  5. —Product Manager, Google

On the record

Humans as pilot · 2025-08

Human judgement, empathy, craft and taste are what it means to be the pilot, not the co-pilot. source

Eval sample quality · 2025-05

Warns against hill-climbing on unrepresentative examples — the eval set has to reflect real users. source

Prototypes over specs · 2025-05

'Prototypes are the new PRD' — he treats built artefacts as the unit of product decision-making. source

Friction in design tools · 2025-12

A single pixel of friction slows down millions of users daily, so AI features must meet Figma's UX bar. source

Interfaces for AI · 2025-12

AI interaction should become more visual and exploratory, and should include teammates, not just the model. source

Interviews & talks

Why it matters here

He is the highest-value single name in this batch. Figma has already decided that design taste must be bought as a graded human service rather than inferred from telemetry, has written the rater guidelines, and is hiring a designer specifically to manage the external partners who supply it. Julian does not have to sell the category to him; he has to beat whoever is currently staffing those nightly evals — and Figma's own posting admits it wants better guidance and quality control over those partners.

How to reach

Open with the eval process itself, not with Miju Labs. Something like: you described taste-and-judgement evals run nightly by contractors against Figma's guidelines; the hard part is rater calibration drift and design-literate raters at that cadence — here is how we staff and calibrate it. Concrete asset to bring: a small worked sample on Figma Make-style outputs, scored on his own 1-4 design/functionality scales with per-rater agreement, so he can compare it against his current vendor's output directly. Speculation: the July 2026 'Product Designer, AI Models' hire is likely now the day-to-day owner of vendor quality, so ask him who that is rather than trying to bypass him. He is highly reachable publicly (X, podcasts, conference stages) and talks openly about the process, which lowers the cost of a cold, specific, non-salesy first message.

What we could not establish
  • Which vendor or vendors currently supply Figma's taste-and-judgement contractors — never named publicly.
  • Whether the July 2026 'Product Designer, AI Models' role was filled, and by whom.
  • His exact current title (sources use Head of AI Products, Head of AI, and Head of Product, AI) and his education.
  • The X handle @DKossnick is presented by search as his account but I could not open the profile to confirm the employer in the bio.

Why they are on the map

Designed Figma Make's human-centric eval process incl. contractor taste evals. source — evidence dated 2025-08

Needs the data for a model; writes the spec.

How Figma buys data

Figma's eval process for Figma Make (First Round, May 2025) used 'taste and judgement' evals by human contractors applying branded guidelines, alongside AI-as-judge and A/B tests. July 2026 job 'Product Designer, AI Models' — 'arbiter of craft and taste', oversees quality of design evaluations and gives guidance to external partners/contractor teams (vendor-managed evals).

AI product org under David Kossnick (Head of AI Products); AI Models team pairs AI researchers with a design-evals designer who manages external eval partners; Weave team from Weavy acquisition.

Known vendors: unnamed external eval partners (per July 2026 job)

Sources

  1. First Round Review — The Art of Evals: how Figma put people at the center of its AI eval process (2025-05-28)
  2. OpenAI — How Figma integrates AI (2025-08-01)
  3. Fast Company — Figma wants to make working with AI more like working with humans (2025-12-04)
  4. Figma job — Product Designer, AI Models (posted 2026-07-22)
  5. Underscore VC — David Kossnick profile (career history)