Somewhere along the way, "will AI take my job?" became the design industry's favourite anxiety. I understand why. I've asked it myself. But after a couple of years of using AI every single working day, I can report from the other side: it didn't take my job. It took my busywork. And I would fight anyone who tried to give the busywork back.
Some context on who's talking: I spent fourteen years in design (illustration, graphic design, web, UI, then UX) before moving into product management in 2022. My design habits never left; they just got a bigger job description. So when I describe what AI does in my week, it's a blend of product work and the design thinking that still runs underneath everything I do.
Here's the honest audit.
What AI actually does for me now
I use Claude daily. Before that it was Microsoft Copilot, which was genuinely useful for documentation but nowhere near as capable. I've also spent plenty of time with ChatGPT and Google's Gemini. The tool matters less than the pattern, and the pattern is that a very specific category of work has quietly left my plate:
Documentation. Briefs, plans, proposals, write-ups, summaries. The stuff that used to eat entire afternoons, not because it was hard, but because it was slow. I still decide what the document needs to say; the machine handles the labour of saying it. First drafts that took hours now take minutes, and my job becomes editing, which is faster and frankly more enjoyable.
Research and analysis. Basic desktop research, competitive scans, pulling patterns out of data. What used to be a browser with forty tabs and a headache is now a conversation. The critical caveat (and we'll come back to this) is that I verify everything that matters. AI research is a brilliant first pass and a dangerous last word.
Prototyping. This is the one that would have blown my 2012 mind. I use AI to build working prototypes of ideas. Not static mockups, actual interactive things I can put in front of people for feedback. Recently I've started using Claude Code to push those prototypes further, with the deliberate goal of shrinking the gap between design and development. The design-to-dev handoff has been the industry's most reliable source of pain for my entire career; watching it compress is genuinely exciting.
Add all of that up and something interesting has happened to the shape of my week. Less production, more direction. Less typing, more deciding. If your identity is wrapped up in the production (and for many designers it is, mine certainly was) that shift feels like a loss at first. It isn't. It's a promotion you didn't have to apply for.
What it can't touch
Now the other pile: the work that AI hasn't taken and, based on everything I've seen, isn't close to taking.
Reading the room. Knowing that the stakeholder who said "looks fine" meant "I have concerns I'm not raising in this meeting". No model attends the meeting-after-the-meeting.
Pushing back on a bad brief. AI is pathologically agreeable. Ask it to design the wrong thing and it will design the wrong thing enthusiastically. Spotting that the request itself is the problem, that the client is asking for a redesign when their actual issue is onboarding, remains a fully human sport.
Accountability. When something ships and it works, or ships and it doesn't, a person owns that. "The AI suggested it" has never once flown as an explanation in any room I've been in, and it never will.
Original judgement about people. UX research, synthesis, the empathy work of understanding why humans do the strange things they do. AI can summarise your research; it cannot have done your research.
The catch: you have to hold its hand
Here's the part the breathless LinkedIn posts skip. AI output almost always needs my input, everything from small tweaks to correcting genuinely large misunderstandings. I've had to remind it of things we'd already agreed in the same piece of work. I've watched it confidently sprint in the wrong direction because my instructions left a door open.
My running joke is that AI is a five-year-old adult. On the surface it's fully formed: articulate, confident, apparently mature. But it's only ever as good as the information you've given it and the guardrails you've set, and sometimes you really do have to hold its hand to get the output you expected. You know the old line about how the smarter someone is, the less common sense they seem to have? AI is the definitive proof.
This isn't a complaint. It's the job description. "Human in the loop" gets used as a reassuring slogan, but in practice it's not a checkbox. It's a skill, and it's the skill. The loop is where the value lives. Every correction I make, every "no, we agreed X, remember?", every catch of a confident-but-wrong claim: that's the part a business is actually paying for. The machine produces; the human is responsible.
And the quality of what you get out is directly proportional to the quality of what you put in. The single biggest lesson from my daily use: the more detailed and focused your instructions, the better the output. Vague prompt, vague result. Treat the lazy prompts like shouting an idea across the office; treat the good prompts like a proper creative brief. Which brings me to the surprise ending of this audit.
The skill that grew in value wasn't a design skill
If you'd asked me in 2022 which of my abilities would appreciate most in the AI era, I'd have guessed something crafty. Visual taste, maybe, or systems thinking. The actual answer: articulation.
The ability to describe intent, constraints, context and taste with precision turns out to be the master skill of working with AI. And here's the kicker: it makes everything else better too. The same clarity that gets a great output from Claude gets a great result from a junior designer, a development team, or a stakeholder meeting. Prompting, when you strip the mystique off it, is just art direction with a keyboard. Designers have been doing the hard part for years; we just called it briefing.
So the busywork is gone, and what's left is the concentrated version of the job: deciding, directing, correcting, owning. That's uncomfortable if you loved the production. It's liberating once you realise the production was never the point.
The machines took my busywork. They left me the work. Turns out those were different things all along. And if you're still not sure where the line sits in your week, try the audit yourself: for five days, note every task and ask "could a well-briefed AI have done this?" The honest answers will tell you exactly what to automate, and exactly what to double down on.
For the concrete version of "what to double down on", read the only 4 AI skills designers actually need.