AI didn't kill product design — it killed the parts that weren't design

There's a tweet I keep seeing variants of: "AI is going to replace product designers in 18 months."

After two years of building real AI-augmented SaaS — shipping Refynes to 500+ Canadian job seekers, integrating Claude into design and engineering workflows daily, building a Figma plugin that turns screens into ClickUp backlogs — I can tell you what's actually happening, and it's more interesting than the doom take.

AI didn't kill product design. It killed the parts of the job that weren't design.

What I used to do that AI does now

Honest accounting. These are the chunks of my week that have meaningfully moved from "me" to "me + Claude":

Generating the first 60% of a screen. Empty Figma canvas, blank <section>. I describe what I want. I get something to react to in 90 seconds. The first draft is no longer the hard part. It used to be a half-day of laying down boxes; now it's a conversation.

Naming things. Variants, components, props, tokens, classes, files. Claude is shockingly good at naming things consistently across a system. I used to lose 20 minutes per session to "what do I call this state."

Boilerplate code. Form validation. A modal. A toast. A reusable hover effect. A sticky scroll observer. I haven't written one of these from scratch in over a year. I describe the behavior, I get the component, I tune it.

Documentation. Every component I ship gets a docstring. Every case study gets a JSON-LD schema. Every commit gets a coherent message. None of that is design work. All of it used to take real time.

Research synthesis. I run user interviews. I paste the transcripts into a project. I ask for themes, contradictions, surprises. The synthesis pass that used to take two days takes two hours.

That's a real list. It's also a list of things that, if I'm honest, weren't the part of product design I trained ten years for.

What AI is bad at — and what that means for designers

Now the part the doomers miss. Here's what Claude still can't do, in 2026, on real client work:

Choosing what to build. AI will happily generate a feature spec for whatever you describe. It will not push back and say "this is the wrong feature." It can't sit in a room with a founder, understand their actual business model, and tell them that their pet idea will sink the product. That's the most valuable design skill, and AI is the least helpful with it.

Knowing when something is "good enough." AI generates beautifully. It also over-generates. Five variants of every button. Eight ways to lay out the same dashboard. Three competing animation systems. A senior designer's job has always been knowing when to stop. AI makes you faster at producing, slower at deciding. That ratio matters more than it used to.

Sensing the room. A clinician using OutcomeMD's dashboard, a homeowner picking a contractor on Blik, a recruiter posting on Refynes — these people have specific anxieties, time pressures, and trust thresholds. AI can simulate a persona. It can't sense the room. Site visits, screen-shares with real users, sitting next to support reps — that work didn't disappear. If anything, it got more valuable, because the people who skip it now have AI to autogenerate "designs" that pass surface inspection.

Crafting a system, not a screen. AI is great at one screen. It's mediocre at five screens that need to feel like the same product. It's actively bad at deciding the underlying primitives — which buttons, which spacing scale, which patterns deserve to become components. That's design system work, and design systems are where most product design value still lives.

Knowing when to break the rules. Every AI-generated design respects every convention. That's why so many AI-only designs feel familiar in a bad way — they look like every landing page, every dashboard. The interesting work breaks one rule on purpose, in service of a specific user moment. AI can copy your taste; it can't develop it for you.

The new shape of the job

If you're a product designer reading this in 2026, here's how I'd describe the shape of the role now:

1. You spend less time producing, more time deciding.

The output funnel is wider. The hard part is filtering. Two hours of generation can fill a week with options. Senior designers earn their salary by killing options fast.

2. You ship more code, or you fall behind the designers who do.

This is the contentious one. I know designers who hate this. I'm telling you what I see on the ground: the designers shipping the most product work in 2026 are the ones who can take a Figma file and turn it into a deployed React component themselves, with Cursor + Claude as the multiplier. Not because they're "10x engineers" — they're not — but because the handoff is gone. Design and engineering are one motion now.

You don't have to become a senior engineer. You have to be able to push a button and see your design as real, working, deployed software. The tools that make this possible (Cursor, v0, Lovable, Bolt) are good enough. The handoff cost is no longer worth the friction it adds.

3. Your craft moves up the stack.

Less pixel-pushing. More product strategy. More taste curation. More writing — yes, writing — because half of designing for AI products is writing (prompts, error messages, model instructions, fallbacks for when the AI is wrong).

I spend more time writing copy than I did three years ago. The interface IS the prompt. The empty state IS the onboarding. Every AI product I've designed lives or dies on how well the human-AI conversation is written.

4. The portfolio bar moves.

I review portfolios for hire weekly. The ones that stand out in 2026 don't have more screens — they have shipped products, written in the designer's own voice, with the specific decisions called out. "We tried X. It failed because Y. We tried Z. Here's what shipped, here's the metric it moved." Generated Figma boards are no longer enough. You need stories AI can't fake because AI wasn't there.

What this means for hiring (because I'm asked a lot)

Companies are confused about what to hire in 2026, so I'll be specific.

The candidates getting offers right now can do all of the following:

• Ship a working prototype the same week as a kickoff. In code, not just Figma.
• Hold a strong opinion in a room of engineers without folding.
• Talk about AI tools like a journeyperson, not like an evangelist or a skeptic. They've shipped with them. They know where each one breaks.
• Articulate their design system principles — not Material, not Apple HIG — theirs. With reasons.
• Show one piece of design writing per quarter. A teardown, a build log, a thread. Public evidence of taste.

The candidates getting passed on:

• Portfolio is all explorations, no shipped product.
• Can describe Figma plugins but can't talk about a real handoff.
• Either dismisses AI ("it'll replace us") or oversells it ("I 10x'd my output"). Both are the same tell: they haven't done the work.
• Hasn't shipped a meaningful product feature in 18+ months.

My current stack

For the curious. This isn't a recommendation, it's a snapshot of what I personally use day-to-day in 2026:

• Design: Figma. Still. The variables + auto-layout system has carried me through every project this year.
• Prototyping: v0 for static hero/page-level concepts, Cursor + Next.js for anything that needs real interaction or data.
• AI agent: Claude (Anthropic), in Cursor, in the API, and in a dozen scripts. The 1M context window is a real difference-maker for codebases the size of a portfolio site.
• Writing: I draft posts like this one in plain Markdown, in Cursor, with Claude as an editor — not a writer. Big difference. The voice has to be mine; the polish can be shared.
• Research: User interviews recorded, transcribed, dumped into Claude projects. Synthesis is one of the cleanest AI wins.
• Motion: GSAP + CSS keyframes. AI doesn't help with motion much yet — taste matters more than code here.
• Code review: Claude reviews every commit before it lands. Sometimes it catches a real bug. Mostly it catches the dumb stuff (typos, missing aria labels, dead branches) so I can focus on the architecture.

Where I think this goes next

Three predictions for 2026–2027, ranked by confidence:

High confidence. The "design engineer" role becomes the modal job title for senior IC product design at startups. The pure-Figma role doesn't disappear — but it stops growing.

Medium confidence. Design systems get smaller, not bigger. The marginal cost of a one-off custom component drops so low (Claude generates it correctly in seconds) that the discipline of "build a primitive everyone uses" weakens. Some teams will benefit from this; most will regret it within 18 months when the design debt hits.

Lower confidence. The portfolio site evolves from "screens" to "products + writing + open source." The case study format I just spent six months perfecting on this site may itself be on the way out, replaced by something more like a developer's GitHub profile. Talk to me in 18 months.

What hasn't changed at all

I'll close on this. Two years into AI-augmented work, with a Claude tab open at all times, with v0 prototypes shipping the week of a kickoff:

The work is still about paying attention to one specific person trying to do one specific thing, and removing the friction between them and the outcome they want. That's it. That's still the whole job.

Everything else around that — the screens, the systems, the code, the documentation — is overhead. AI shrinks the overhead. It doesn't shrink the work.

The designers who'll do well in the next decade are the ones who recognize which is which.

---

If you're hiring for senior or staff product design in Canada or remote North America — design + code, AI fluency, healthcare / fintech / retail / SaaS experience — I'm open to short conversations. Email is on the about page.

If you're a designer trying to figure out which AI tools are worth your time, I usually answer DMs on LinkedIn.