The short version: On July 23, Paper closed a $34M Series A led by Accel and ICONIQ (total funding now ~$38.5M), with WorkOS's Michael Grinich, Lovable's Anton Osika, and folks from Anthropic and OpenAI writing angel checks. It says ARR grew 25x since launching Paper Desktop in early 2026. But the number a founder should copy from this round isn't the valuation — it's the format decision underneath it: Paper's designs render in real HTML and CSS, so humans and AI agents edit the same artifact that ships.

What Paper is#

Paper Desktop is a shared workspace where design, code, and data live in one document that updates in real time — for both humans and AI agents. The customers on record are the kind of teams that live in coding agents all day: Ramp, Lovable, Vercel, PostHog, Quartr, Y Combinator. Four straight months on Ramp's Fastest Growing Companies list, 25x ARR since launch. Those are the metrics that got Accel and ICONIQ to lead.

We covered the product itself when it first showed up in the Paper design-platform tool highlight. This piece is about why the round validates a specific technical bet — one that generalizes past design tools.

The bet: your design artifact should be your production artifact#

Classic design tools store your work in a proprietary vector document. That's fine when the next step is a human engineer who re-implements the frame in code. It breaks the moment the "engineer" is an AI agent.

Ask a coding agent to "build this Figma frame" and you've handed it a picture. It has to infer the layout, guess the spacing, reconstruct the component tree, and re-type it as code — and every inference is a place fidelity leaks. The design-to-code handoff was always lossy; agents make the loss expensive because they do it at volume and confidently.

Paper's answer is to delete the translation. If the design is already HTML and CSS, there's no reinterpretation step — the agent reads and edits the exact artifact that goes to production. Designer and coding agent are in the same file, editing the same boxes, in the same format the browser will render. No export. No "does this match the mock." The mock is the code.

The rule for founders, whatever you build#

The $34M is Paper's. The transferable insight is cheaper than a Series A:

Give your agents artifacts in their production format — not a rendering of one.

Everywhere your workflow hands an agent a lossy representation, you've planted a failure. A screenshot instead of the DOM. A PDF instead of the structured data. A design image instead of the markup. Each forces the agent to reconstruct what you already had in exact form, and reconstruction is where agents hallucinate. Paper's whole thesis is that the design tool of the agentic era is the one whose native format an agent doesn't have to decode.

So the audit is short: look at every point where a human or an agent hands work to the next step, and ask whether the handoff is the real artifact or a picture of it. Where it's a picture, that's your next agent bug — and, as Paper's round suggests, possibly your next product.

The pattern rhymes with the other bets we've watched get funded this month: valuations are increasingly priced on a nameable, load-bearing reality rather than a demo — the committed offtake contract behind Humanoid's $1.35B, the 95%-small-model production mix behind Fireworks. Paper's is the same shape: it raised on a format choice that removes an agent's most reliable failure, and it has the 25x curve to show the choice was right.