Fast, fast, fast, then slow: the wall AI building hits last
One line in Figma's Build by design launch post describes building with AI better than anything I've read this year: "It's fast, fast, fast... then slow." I've felt that shape all year — an afternoon that produces a month of candidate screens, then days of squinting at them. I wrote in Why curation is the job now that deciding is the job and generation is the cheap half. This week, that argument got numbers attached.
Figma ran a randomized controlled trial on its own tool and published the inconvenient half: the people the tool was built for barely sped up, and the people who used to wait on them got most of the gain. The slow part after the fast part is now most of the work.
Why is one day of building fast and the next month slow?
Because generation has no opinions and no accountability. The model optimizes for plausible. It has no stake in whether a screen should exist, which of four candidate dashboards deserves the users, or what a feature is called. Everything after the first prompt is naming, hierarchy, and the should-it-exist call — judgment work, and judgment has no autocomplete.
On a product team the slow part looks like review threads and a calendar. On a solo product it looks like you at midnight with every version still open. Same wall, different furniture. The difference this week is that the company selling the fast part published numbers showing their buyers hit the wall too.
What did the Figma Make trial measure?
Figma paired with the research firm MeasuringU and posted a randomized controlled trial on arXiv on September 22: 50 product designers, 50 product managers, three standardized UI editing tasks, half with Figma Make and half without.
The rank order is the headline. PMs saved 22 minutes 46 seconds overall — 35% — and their biggest win was 39% on the smallest request, adding one item to a settings menu. Designers only saved real time on the hardest task, a comment flyout, 26% faster. Across all three tasks their gain was 17%, right at the edge of significance, and the paper's own discussion says it doesn't hold up as significant. Quality was never scored against users; the study times how fast you reproduce a design someone already decided on.
Figma's blog summary of the Figma Make time-savings study puts it plainly: a PM with Make was almost as efficient as a designer without it. Execution got cheap for almost everyone. Judgment didn't get cheaper for anyone.
What shipped to speed up the slow half?
Figma's answer to its own data landed October 15, and most of it touches the decision, not the generation:
- Code layers went to beta: code-backed screens you explore on the canvas with a team. That replaces the single-player tunnel of a coded prototype, where everyone else only saw the end result and never the path.
- Specs gives decisions a home in open beta. Diagrams, specs, and canvas work sit in one place, so the reasoning survives the handoff and travels to a coding agent through MCP.
- Design system guidelines attach to published libraries, so agents build on the patterns a team already decided. A system map and system check keep that stack honest over time, with twenty MCP partner integrations wired to the tools agents already run in.
That is infrastructure for the slow half. The quote everyone shared was the symptom; these three are the product roadmap admitting it.
What does a solo founder do with a slow week?
You can't out-review a team, so you shrink the surface the review runs on. Three things I keep doing:
Defaults before prompting. My tokens, components, and type scale are set before any generation runs, so the agent inherits decisions instead of inventing them. Small batches, one sitting — my keep rate across outputs stays under a third, which matches what I measured when curation became the job I actually do. And one written line per survivor: why this one lived. Next week's prompts read that log.
None of this makes generation faster. It makes Friday's review shorter — two candidates to argue with instead of ten. For a solo builder that hour is the product. The model already ate the other hours.
Frequently asked questions
A randomized controlled trial with 100 participants found roughly 20% faster overall completion with Figma Make. Product managers saved 35% cumulatively, including 39% on the smallest task. Product designers showed no significant overall time saving — their one clear win was 26% faster on the hardest interaction prototype.
About the author
mosh
mosh is a product designer for growth, working with design thinking and ever-improving design systems. What matters: fixing conversion, whether in B2B dashboards or direct-consumer apps.
Keep reading
- Escape Ludo: game design
A covid quarantine project. Sprite design, character formation, levels, and inventory UI for a multiplayer Ludo game.
- Go Laundry: service booking platform
One platform to book cleaning services in Qatar. Art-directed brand mascots, app UI, and UX research.
- SaaS activity tracking: designing an admin log teams trust
Admins needed to see every team action in one place. Research with two managers and eight user stories turned a confusing log into filters, search, share, and export.