Design Critique

Design critique is the surviving core of senior design work after AI ate four of the five traditional workstreams. It is the act of applying taste, strategic judgment, and decision making in areas of genuine ambiguity that no prompt can fully resolve yet. The old version at Google, Meta, Airbnb, and Atlassian in 2023 through 2024 involved a senior PD owning the Figma file, running hour long review sessions, redlining junior mocks live, capturing sticky notes, and producing long synthesis docs. That ritual died in 2025. The new version assumes tools like Cursor, v0, or Claude Code already generated the first pass using the complete brand brief, every logged critique from the past year, the live component library, and tagged Linear tickets. The senior now focuses solely on the calls AI cannot own: whether the new illustration style at Vercel feels too playful for their technical audience, or if the motion curve in Linear's command bar would annoy power users expecting zero latency. This matches the autopsy in The Death of the Senior Designer. Research synthesis became a Linear AI summary. Spec writing became a Claude export. Mock polishing became Figma AI plus v0. The first pass at critique itself got automated. What remains is the reduced, sharper layer of taste plus ownership that the shipping senior exercises right before merging the PR.

Design critique is not group therapy or design by committee. It is not the 12 person Zoom where PMs throw requirements, engineers push back on every radius, and everyone leaves with 47 Figma comments that dilute the work into mediocrity. It is not the polite have you considered thread or the junior defending pixels for 45 minutes. Those were always theater. They scaled poorly and delivered average outputs. AI exposed the waste. When a single prompt can ingest prior critiques and return tighter analysis than most mid level designers, the performance aspect of old critique becomes indefensible. Real critique was never democratic. It was one or two people with skin in the game making hard calls on trust, clarity, or emotional tone. The new version strips away the theater and keeps only that decisive core. Everything else got commoditized by tools like Lovable, Galileo, Magician, and Subframe.

Concrete example one. Early 2026 at Vercel the team shipping updates to v0 ran critique on a new template gallery surface. The design engineer loaded a prompt into Cursor containing the 2025 brand audit, all six prior critiques on similar surfaces, the Tailwind config from the live codebase, and user feedback from the last Linear cycle. Claude returned a structured 1100 word critique calling out overcrowding in the preview cards, insufficient contrast on hover states for WCAG compliance, a motion curve that felt sluggish compared to Vercel's speed brand, and a recommendation to align the illustration style with their 2025 campaign assets. The senior reviewed it against the deployed preview URL in under eight minutes. He accepted the contrast and spacing calls, overruled the motion suggestion because slowing it down would contradict the speed is the brand principle, and adjusted two illustration weights directly in code. The PR merged the same day. No Figma file existed. No handoff spec was written. Total critique time sat at 37 minutes from prompt to production. The old 2023 version of this at the same company would have burned two weeks on mocks, a 90 minute stakeholder meeting, and three revision cycles before engineering saw it.

Concrete example two. At Anthropic on claude.ai updates the contextual critique runs even tighter because the domain carries regulatory weight. The prompt library includes the full accessibility spec, taste principles locked in at the 2024 offsite, every prior critique on tone of voice, and legal exposure notes around perceived accuracy. When a new helpful suggestions panel was generated, Claude flagged four issues including overly warm language that risked reducing perceived precision for technical users, hierarchy problems in the new AI toggle, and spacing inconsistencies with the marketing site. The design engineer who owns the surface made the final two calls: keep the warmer tone in secondary areas only and increase visual separation on the toggle to prevent misclicks. Those decisions shipped immediately. This is the deciding senior at work. Compare it to a traditional senior PD at a Series B SaaS in 2024 who spent days prepping for critique, synthesized research manually, polished in Figma, then handed off to engineers. The new loop delivers three to five times the surface area at identical taste level.

Concrete example three. A freelance design lead who retrained in 2025 used the exact playbook. She picked React and Tailwind, built one fully shipped product using Cursor as daily driver, replaced her case study deck with a live URL and GitHub repo, and now quotes 50 percent above her old senior PD day rate. Her critique skill transferred perfectly. Clients like founding teams at companies similar to Lovable feed her their prompt library and Skill packs. She runs the final taste filter on deployed surfaces, makes the ambiguous calls, and the work ships. No more 200 comment Figma threads. The artifact proves the shift.

Use the new design critique when AI has already generated surfaces and you need a final human filter on high stakes taste, brand trust, or strategic differentiation. Run it at pace setting teams like Linear, Vercel, Stripe, Anthropic, or the Browser Company where critique happens against GitHub PRs and live URLs instead of static mocks. Use it if you have retrained into the shipping senior or design engineer role that owns the full loop. Use it in the four narrow surviving seats for traditional seniors: leadership at scale above 500 designers, pure long cycle research, regulated high liability domains like healthcare and finance, or deep accessibility specialism that carries legal weight. Do not run the old version with large meetings and Figma files. That model died with the handoff. Do not use it for synthesis, research clustering, or polish. Linear AI, Notion AI, v0, and Claude handle those at L5 quality. Do not use it if your company still hires traditional senior product designers who only spec and critique without shipping code. You sit in the lagging cohort headed for the next reorg. Run the five question audit from the parent article first. If three answers point toward Figma and specs instead of PRs and shipped URLs, you have twelve months to move.

Design critique survived AI by shrinking to the one thing only a shipping senior can truly own: the final taste call before the PR merges.