The creative tool layer

If the frontier labs are the prestige customer, the creative app layer is supposed to be the realistic one: dozens of companies, fresh capital, design quality visibly gating their product, and no in-house research organisation to build around. That is the hypothesis. It does not survive the evidence.

These companies employ designers in volume, at $105K–$376K. They do not appear to buy designer judgement.

#$31.6B of valuation, zero evaluation headcount

Between them, roughly $31.6B of valuation and nobody whose job is to evaluate design quality. UI-Bench ranks all three, and Replit sits at the bottom of it — a measured, public, unaddressed design-quality problem at a $9B company.

There is corroborating signal that the question is live: Business Insider covered "Can AI have taste?" as a hot topic at Replit's New York vibe-coding conference in June 2026. But interest at a conference is not a budget line, and no budget line is visible.

The most plausible reading is that these companies are growing so fast on distribution and model access that design quality is not yet the binding constraint. UI-Bench suggests it will become one — it exists precisely because outputs differ in design quality despite sharing underlying models. That is a timing argument, not a demand argument, and a client should price it as such.

#Figma is pure build, and the eventual competitor

163 postings, 19 of them design, bands $105K–$376K, including Product Designer, AI Models and Product Designer, Design, Dev & AI Tools. Zero evaluation, human-data, annotation or rater roles. Its Designer Advocate roles at $153K–$317K are developer relations for designers, not evaluation (Greenhouse).

Figma is a company whose entire staff is a design-taste asset. It is the least likely buyer of external design judgement in the landscape, and the most likely eventual supplier of it.

#Canva is the exception and says so in writing

Canva is the only app-layer company with a visible, funded, explicitly named evaluation function. Its Principal Research Scientist, Evaluations posting is the single most important document in the buyer research:

And, decisively for procurement:

"Human data spend" is a named budget line at Canva. It is the only explicit confirmation of a human-data budget at any app-layer company in this research. The role is company-wide, "setting direction across our research groups in Australia, Europe, the US and China," covering design, image, video, audio and agentic workflows. Canva runs four other evaluation-adjacent roles: a second Principal Research Scientist for Evaluations in San Francisco, an Engineering Manager (ML) for the Evaluation Platform, and a Senior Research Scientist for Design Generation.

#But Canva's supply answer is already built

The same board discloses how Canva sources the labour. Its AI Quality Evaluator (Dutch, 12-month contract, Amsterdam) is tasked with:

So Canva has in-house labelling teams in the Philippines and the EU, per-language contract evaluators hired directly, and a Principal-level scientist allocating the budget between compute and human data. That is a sophisticated build, not a procurement gap.

One caution on timing: Canva's valuation was cut by roughly $10–11B in August 2026 from about $42B, against roughly $4B ARR, "putting IPO plans in doubt". A company defending a valuation cut is a harder sell for new discretionary spend and an easier sell for anything framed as replacing in-house headcount. Frame accordingly.

#Buy versus build, company by company

Boards returning zero creative or evaluation roles: Webflow (27), Descript (8), Ideogram (3), Pika (10), Hedra (8).

#The Contra partner list could not be corroborated or refuted

The claim under test is that Contra Labs' only corroborated partner list is Framer, Webflow, Lovable, Replit and HeyGen, sourced from a launch newsletter.

• contra.com/labs and contra.com/ai both return 404. There is no live partner page at the obvious URLs.
• None of Framer, Webflow, Lovable, Replit or HeyGen mentions Contra, Contra Labs, or any external evaluation vendor across their 176 combined machine-readable postings.
• Webflow provides no signal either way; Framer and HeyGen have no public board at all.

So the list stays [UNVERIFIED]. What is corroborated is Contra Labs' existence, scale claim and pricing, via Fast Company on 24 August 2026: an "independent human data and creative evaluation lab launched by hiring platform Contra" connecting "more than 1.7 million creative professionals" with AI developers, where creatives can earn "from around $50 to $250 an hour working on RLHF and other types of post-training" (Fast Company).

Note the asymmetry in that rate: it is what workers are paid, which implies a materially higher price charged to buyers — typically 1.5–2.5× — but no margin or contract value could be established. GMV is not revenue applies.

#What the silence actually means

Zero vendor mentions in 1,562 postings admits two readings, and honesty requires holding both.

Reading one: they are not buying. These companies employ designers, ship fast, and treat design quality as a staffing question. Every signal in the table above is a build signal.

Reading two: purchasing never touches a job description. This is common — procurement rarely surfaces in ATS text, and the same search across the frontier labs found no vendor names either, in a sector we know buys through vendors because OpenAI's own postings describe managing them. See The generative-media labs.

The methodologically defensible conclusion is narrower than either: there is no visible evidence of app-layer demand for external design evaluation, and the absence would look identical if the demand existed and was procured quietly. What tips it toward reading one is the corroborating structure — no evaluation headcount either, which purchasing would not hide. A company that buys external evaluation still needs someone internal to specify and consume it, and that person is not being hired anywhere except Canva.