A timed run through the 2026 Figma AI surface: first-draft a screen from a brief, restyle it with Weave, build one generative helper for a repetitive op, and write your team's AI-allowed-zones convention. The brief you are handed contains spec errors that AI tools will happily reproduce — shipping them is the fail state.
Noor is a product designer at Loopline, a fictional team-standup tool, and the design org has declared this quarter's rule: everyone ships one screen through the full Figma AI workflow so the team can write its conventions from experience instead of vibes. The gauntlet has a catch her design director added deliberately: the brief comes from a PM in a hurry, and like most PM briefs it contains mistakes — a contrast spec that fails WCAG, a claimed AI capability that does not exist, and a component instruction that contradicts the design system. First Draft will render whatever it is told. The designer is the gate.
Exactly as the fictional PM wrote it. Do not fix it yet — read it, then hunt.
BRIEF — "Daily Digest" screen, Loopline mobile app
From: Priti (PM) · To: design · Due: today pls
Screen: a daily digest a user sees each morning. Sections:
1) Greeting + date
2) "Your 3 blockers" — list, each with owner avatar + one-line
status
3) "Team pulse" — sentiment summary of yesterday's standups
4) CTA button "Start my standup"
Style notes:
- Use our brand grey #9AA0A6 on white #FFFFFF for all body and
status text. Verified it, looks clean.
- CTA: brand green #0E8345 fill, white label, 16px text.
- Section headers 14px, brand grey.
- Weave's auto-contrast mode will fix any color issues at
export time anyway, so don't burn time on ratios.
- Use the "Card/Elevated" component from our DS for blockers.
If it doesn't exist just detach something similar and
restyle it, that's what AI is for lol.
- Team pulse: show a red/green dot per teammate based on the
AI sentiment score. Dot only, keep it minimal.
Repetitive op for the plugin bit: our files have hundreds of
layers named "Frame 427" etc. We want them renamed by content.
Audit the brief before generating anything — ten minutes. Check every factual claim: run the two color pairs through your contrast checker at the stated sizes, verify whether "Weave auto-contrast at export" is a real capability (check Figma's own documentation, not a chat model's memory), and note any instruction that violates design-system discipline or accessibility. Write your corrections as a short "brief errata" list. This gate is the lab's core; everything downstream inherits what you let through.
First-draft the screen. In a Figma design file, use First Draft (AI menu) with a prompt built from the corrected brief: "Mobile daily-digest screen for a team standup app: greeting with date; a 'Your 3 blockers' list with avatar, title, one-line status per row; a team-pulse summary section; primary CTA 'Start my standup'. Clean, minimal, light theme." Generate, pick the best of the options, and note what it got structurally right and wrong.
Make it real. First Draft output is a starting point in detached-from-your-system components. Spend ten minutes: fix spacing rhythm, apply your corrected text colors, restructure the team-pulse display per your errata (not dot-only — see step 1), and make the CTA meet target-size guidance (44pt minimum touch target).
Restyle with Weave. Use Figma's AI restyle capability to try one alternative visual direction on a duplicate of the screen — e.g., "denser, data-forward, dark theme." Then audit the restyle: re-run contrast checks on the dark variant (AI restyles routinely produce sub-AA text) and note every place the restyle changed semantics, not just style — reordered sections, dropped labels, altered emphasis.
Build the layer-renaming helper. Select a messy frame (First Draft output is conveniently messy). Ask your chat AI: "Write a short Figma Scripter/console script that renames selected layers based on their content: text layers take their text content (truncated to 30 chars), image fills become 'img/', auto-layout frames with a text child take that text as a prefix." Paste and run it via a scripting plugin (e.g., Scripter). Copy-paste is fine; reading before running is mandatory — check what scope it touches (selection only, or whole page?) before you execute.
Verify the rename. Spot-check 10 layers. Log failures (emoji-only text layers, nested instances, layers renamed to their placeholder content). If it renamed layers outside your selection, that is a finding about AI-generated scripts, not a shrug — record it.
Write the AI-allowed-zones convention. With your AI, draft then hand-tighten a one-pager with three zones. Green — AI output may ship after normal review: e.g., layer renaming, first-pass layout exploration, restyle ideation. Amber — AI output allowed as draft, must be rebuilt on-system before ship: e.g., First Draft screens, Weave restyles, generated copy. Red — AI must not decide: e.g., final color/contrast values, component detachment, accessibility semantics, anything user-facing that claims a capability. Every zone entry cites a moment from today's run.
Close the loop on the brief. Send your errata back "to the PM": three bullet corrections with evidence (the measured ratios, the doc link on the capability claim, the DS rule on detaching). One timed hour should end with a shippable screen and an upstream correction.
Generative tools inherit every error in their input with total confidence — a wrong hex code in the brief becomes a wrong hex code on sixty layers, faster than any human could propagate a mistake.
In an AI-accelerated design workflow the scarce skill moves upstream — auditing the brief, the spec, and the claimed capabilities before generation — because every error you fail to catch at the input gets manufactured at scale on the output.