AI is changing product development faster than job titles can keep up with.
- Designers are working closer to code.
- Engineers are shaping the experience earlier.
- Product managers are prototyping instead of waiting for a handoff.
The work hasn't disappeared. Its boundaries have. So has its tempo.
Take Val; she is a designer who needs to ship a new onboarding flow. A year ago that meant research, a round of mockups, a review, a handoff, a wait. Today she might be adjusting real interaction states in vibe code before the mockup is even finished, because the working version got there first. Same job. A completely different shape. We'll come back to her.
Three recent sources point at the same shift from different angles: Jenny Wen's account of design at Claude, Figma's research on shifting product roles, the McKinsey Global Institute's read on people-AI collaboration. Together they land on something practical: AI will automate parts of the work. And the bigger change is how teams organise around it.
The linear process is giving way to a loop
The classic sequence looks tidy in a common narrative:
- Ideate
- Sketch wireframe
- Get feedback
- Iterate
- Finalise pixel-perfect UI
- Hand off to engineering
Real teams never followed that neatly. AI-assisted artefacts rewrite the gap between the narrative and reality and make it impossible to ignore.
On Lenny's Podcast, Jenny Wen, who leads design at Claude, describes a workflow where engineers produce working versions so fast that design can't survive as a long phase ending in a handoff. Designers join the loop early instead. React to real model behaviour. Refine the thing as it's being built, not after.
That changes what counts as a useful artefact. Our designer's pixel-perfect mockup still earns its place when precision is the point — but it no longer gets automatic status as the output of design. A rough prototype. A working interaction. A short written direction. Any of these matter more if it facilitates collaboration and ships something sooner.
The new rhythm is:
- Build
- Evaluate
- Iterate
Research still happens when it answers a real question, not because a checklist says the next phase has begun. Process stops being a sequence. It becomes a toolbox.
Design work is splitting into two modes
Wen separates the designer's contribution into two.
-
One mode helps engineers ship fast. Shortening feedback loops and deciding when something's good enough to test, instead of waiting for a formal review.
-
The other provides direction. When a team can generate a dozen plausible features by lunch, someone still needs a coherent view of where the experience should go, even if the horizon has become shorter.
A three-to-six-month direction can guide an AI product better than a five-year vision ever could, simply because the ground underneath moves faster than distant detail stays true.
Our designer still has to decide which version of the flow is actually coherent, and which one just looks finished.
Both modes rely on the same thing: judgement. Speed creates more options, not more clarity.
Role boundaries are already dissolving
Not just in design. Figma's 2025 research (51 interviews and a survey of 1,199 product builders) found most respondents identify with more than one role. PMs sketch interfaces. Engineers prototype interactions, and designers work inside the build tool. Titles that used to mean one thing now cover two or three.
That overlap isn't a side effect of better tools. It's becoming the operating model itself. Most of Figma's respondents named AI as the main force behind the shift. Most called it a net gain: more time on the work that matters, more of it done. But faster output brings its own weight. More tools to learn. More generated material to sort through. More decisions about where a human's attention is actually worth spending.
The clearest casualty is the handoff. When a designer and an engineer can shape the same prototype side by side, waiting for one to finish before the other starts adds delay and nothing else. Val, our onboarding-flow designer, isn't handing anything off anymore; she's building next to the engineer, in the same file, at the same time.
Share of working skills
McKinsey Global Institute estimates current technology could, in principle, automate activities that cover the majority of US work hours. Capability, not adoption. Not job losses. Cost, regulation, trust, and plain business judgement decide the rest.
Most of today's skills already live in work people do together with AI. A small share sits mainly with AI. A similarly small share stays mainly human. Nothing vanishes — the skills just move house.
Take problem solving. An agent can review data, spot a pattern, propose an option. A person still has to frame the problem, question the assumptions, decide what happens next. That reframes the question worth asking. Not: “Can AI produce this screen?” But: “Across the whole path from customer need to shipped product, where should AI accelerate the loop, and where does a person still have to supply context, care, accountability, all the things a model can't?”.
Jobs will change in both directions
Employment forecasts deserve the same caution. The World Economic Forum's Future of Jobs Report 2025 projects tens of millions of roles displaced by 2030, and a larger number created. Meanwhile, what each role requires shifts underneath it.
No one should mistake a global forecast for a personal guarantee. But the numbers do rule out the simple replacement story. Some work disappears. Some expands. New combinations show up that were never a job title before. The real response isn't panic or comfort; it's building the skills that get more valuable as routine production gets cheaper and faster.
What stays distinctly valuable
Quality is still the differentiator. AI can produce a plausible first pass in seconds, which is exactly why generic output is suddenly everywhere. Craft lives in what happens after: the sharper hierarchy, the edge case someone actually thought through, a product that feels intended rather than assembled.
That's the whole distance between Val's first draft, designed with AI, and the onboarding flow she actually ships.
Human review matters more as volume rises, not less. Someone still has to catch the error. Notice when the model is running without real context. Decide when a polished-looking answer is quietly wrong. At the same time, stop practising that supervision, understanding when something is good enough goes with it.
Vision is a constraint on speed, not a brake against it. Without direction, a team can ship more features and still end up with a fragmented product. A shorter horizon doesn't kill strategy, but revisits it constantly instead of writing it once and following blindly.
Empathy, communication, and collaboration stay valued for the plainest reason: cross-functional work runs on shared understanding, and AI commercial models can't manufacture that. They can help produce more, but they can't settle what the user problem is. They can't decide who owns the final call.
How product teams can respond now
Start by treating the process as a toolbox, not a sequence. Keep whatever step actually reduces risk or builds alignment. Drop whatever step survives only because the old process made it standard. Then invest in evaluation itself. Practice why one output beats another, build a shared sense of what good looks like, because that's the difference between fast generation and useful work.
That means working across the old lines on purpose.
-
Designers and PMs gain from enough build-tool fluency to stay close to implementation.
-
Engineers gain from treating interaction quality as part of the product, not a layer bolted on after.
Everyone gains from sharing rough work early: an unfinished thing exposes an assumption faster than a polished one ever does, and shows exactly where the model can be trusted and where it can't.
The least comfortable part: study the failures, not just the capability. Knowing when to trust an output, when to override it, how to check it. That's the whole gap between a team using the tool with confidence and one just collecting prompts.
The role is expanding
Back to where we started.
- The designer working closer to code.
- The engineer shaping experience earlier.
- The PM prototyping instead of waiting.
None of that is a temporary blur while the tools catch up. It's the new shape of the job, already settled in.
AI lowers the cost of making something. It does nothing to lower the cost of deciding what's worth making, whether it actually works, how the pieces fit. Those decisions are the job now.
Val, our designer, didn't ship the onboarding flow because AI built it for her. She shipped it because, out of everything AI could generate, she knew which version was right.
Sources
- Figma Report: https://www.figma.com/blog/2025-shifting-roles-report/
- McKinsey Global Institute Research: https://www.mckinsey.com/mgi/our-research/agents-robots-and-us-skill-partnerships-in-the-age-of-ai
- World Economic Forum Report: https://www.weforum.org/publications/the-future-of-jobs-report-2025/digest/
- Jemmy Wen hosted in Lenny's Podcast: https://www.youtube.com/watch?v=eh8bcBIAAFo