Product design is the first role in the v1 capability matrix where the share-weighted picture has a non-trivial Replaceable wedge — but the role itself isn't replaceable. The two facts coexist because design decomposes cleanly into work the model handles and work it cannot, and the Replaceable wedge sits exactly on the visual-asset generation cell that has been the most visible AI-displaced surface since 2023.
This post reads the six representative tasks from the v1 capability matrix for Product Designer and lands the share-weighted picture for a typical Tier-2-mid cell.
Task-level read
Generate visual assets. Capability 85, reliability 80, error cost 1, oversight 10 min. Classified Replaceable. This is the only Replaceable task across all 15 v1 roles where the capability score is above 80 and the reliability score is above 75. Frontier image models produce production-quality marketing visuals, icon variations, and asset families for design systems. Error cost is the lowest in the role because a wrong visual is cheap to redo, not a load-bearing artifact.
The honest framing: this used to be 10–25% of a working designer's hours. By 2026, it's closer to 5% — and most of that 5% is curation, not generation. The economic substitution here is real and already priced into design-team headcount in companies that have updated their staffing model.
Design wireframes / hi-fi mockups. Cap 65, rel 55, err 2, oversight 30 min. AI-augmented. Frontier models can rough-in screens from a verbal brief, especially against established design systems. The cap is held below the visual-asset cell because mockups carry product judgment — what to put on the screen and what to leave off. Reliability 55 captures that "looks right" and "is right" diverge here in a way they don't for visual assets.
Design QA + handoff. Cap 65, rel 60, err 2, oversight 25 min. AI-augmented. The work of producing handoff specs, redline annotations, and design tokens documentation is structurally a transformation problem. Models handle it well. The oversight is checking that nothing was dropped, not generating from scratch.
Design system maintenance. Cap 55, rel 50, err 3, oversight 40 min. Human-led, AI-assisted. AI can propose component variants and flag inconsistencies across a Figma library, but the judgment of "what should this component be" stays with the design system owner. Reliability 50 is the gate: AI-proposed system changes often look right and create subtle divergence that compounds across hundreds of downstream uses.
Conduct user research. Cap 30, rel 35, err 3, oversight 75 min. Human-critical. AI is useful for synthesizing transcript piles (where the input is text and the output is themes) but cannot run the interview, hold rapport with a stranger, or recognize when the participant just said something contradicted by their behavior three minutes earlier. The 75-minute oversight pass is the analyst pulling apart the AI's synthesis against the source recordings. Cap 30 captures the limited surface where AI helps — not the bulk of the work.
Stakeholder design reviews. Cap 25, rel 25, err 3, oversight 60 min. Human-critical. This is the cell where designers argue against PM-requested patterns that compromise the system, push back on dev pushback when the system is right, and read the room for "what does leadership actually want here." Same shape as engineering and PM stakeholder cells — low cap, low rel, expensive when wrong.
Share-weighted summary
For a typical Tier-2-mid Product Designer averaging standard task-hour distribution: ~10% Replaceable (visual assets), ~35% AI-augmented (wireframes + handoff), ~25% Human-led-AI-assisted (design system maintenance), ~30% Human-critical (user research + stakeholder reviews).
The 10% Replaceable wedge is what makes this role read as "AI displaced" in popular framings. It's real — that 10% of hours has materially compressed. But the share-weighted picture says the other 90% is intact, and 30% of it is gaining importance as AI accelerates everything around it.
Operational AI cost for the AI-augmented portion runs $2,400–$3,100 per month at typical task volume. Against a $130K fully-loaded annual salary, that's a cost ratio of one-to-three on the substitutable portion — same shape as engineering and data work. The Replaceable wedge ($800–$1,100/month of AI cost replacing roughly $1,500/month of designer time) is the only place AI is straightforwardly cheaper.
The visual-asset story is overweighted
Trade media has done a thorough job documenting individual designers losing visual-asset work to AI. Those stories are accurate. They're also a 10% wedge of the role, not the role.
The interesting story is what designers do with the 10% of recovered time. Two patterns from companies that have updated workflow:
- Time shifts into user research — the cell with the highest Human-critical valence and the longest queue at most companies.
- Time shifts into design system maintenance — the cell where AI is useful enough to be tempting but unreliable enough to need human judgment as the final arbiter.
Both are upgrades from "produce icon variants for the marketing team."
What to do with this
Capture the visual-asset frontier. If your design team still bills hours against icon variants, marketing collateral, or asset family expansion, those hours should be 70%+ on AI tooling by end of 2026. The economics check out — and resisting it costs design morale, not just budget.
Don't outsource the design system. Capability 55 / reliability 50 on design system maintenance is the right number — useful, not reliable. The pattern that breaks design systems is AI-proposed components that look correct and create downstream inconsistency. Keep humans on the final-call.
Re-allocate research hours up. The Human-critical cells in this role are research and stakeholder reviews. Both compound — the better your research, the more leverage every other design hour generates. Cap 30 means AI is a junior assistant on synthesis, not a substitute for talking to users.
See the /roles/product-designer single-cell read for the Tier-2-mid breakdown, /insights/product-designer for the cross-cell distribution as Wagecards accumulate, and /methodology for capability-score derivation.