I recently watched a video about scheduled agents in Claude Code. You point an agent at a repo, give it a prompt and a schedule, and it runs that task on repeat in the cloud. Every hour, every morning, whatever you want. The person in the video had an agent monitoring Sentry and auto-fixing bugs. Another auditing dependencies. Another updating docs. He set them up, closed his laptop, went to bed. By morning there were six PRs waiting for him. Each one described the issue it found, explained the root cause, and proposed a fix.
All of this happened while he was asleep.
This isn't one person's experiment either. According to Steve Kaliski, a Stripe engineer, they're already merging 1,300 agent-initiated PR's per week with no human involvement beyond review (source). The factory isn't coming. It's running at full steam.
The framing in the video was around engineering... pipelines, tokens, CI failures. He called this the "software factory" model. Agents on the assembly line, humans as floor managers.
As I kept watching, all I could think was: where's the designer in all of this? When engineers can ship at this pace, how the hell do we keep up?
The job market is already answering that question, and the answer isn't comforting. Engineering roles are at their highest in three years. PM roles too. AI-specific engineering and PM positions are up over 400% from their lows. And design? Plateaued. Barely up from where it bottomed out in 2023. Companies are investing in people who can build and people who can decide what to build. The traditional design layer is flatlining.
I wrote recently about design engineering as the next era of product design. Learning the medium, building directly, closing the gap between intent and output. I still think that's the right move for right now. Dev skills give designers the technical fluency to work with AI, build directly, and understand what they're shipping. But how long that window lasts is the open question. Maybe a year?
Then what?
Look at where the demand is growing. It's not design execution. It's product judgment. Understanding users deeply. Deciding what to build and what to cut. Connecting business goals to experience decisions. The market is repricing who gets to own the whole problem. That's product thinking, and the data suggests it's where the value is moving. Even engineers are seeing it. As Steve Kaliski from Stripe put it: if coding becomes free, the bottleneck shifts to having enough good ideas in the first place.
AI is collapsing the skill gap on the execution side for everyone. When anyone can build, the differentiator stops being who can design or code faster. It becomes who can think across the entire product. Designers have a specific advantage here. Product thinking isn't just prioritisation and roadmaps. It's watching someone use a product and understanding why they hesitated. What they expected that didn't happen. Where the friction actually lives. Designers have been building that muscle for years. They've just been applying it to screens instead of strategy.
In an agentic workflow, that judgment gets expressed differently. You're not designing screens. You're setting the conditions that agents build within. Writing specs precise enough to build from but flexible enough for reasonable decisions. Defining policies for how the product should behave... when to use progressive disclosure, what an empty state should communicate, how destructive actions should be confirmed. Those rules currently live in a designer's head. They need to become externalised and testable. Something closer to a test suite than a style guide.
The stakes are higher than they sound. If the spec was vague and the agent builds something technically compliant but experientially wrong, that's on the human who wrote the spec. Accountability moves upstream. The person who defined the intent owns the outcome.
And then there's the most difficult part to encode: the gestalt. The overall feeling of moving through the product. The pacing of information. The moments where the interface is quiet versus where it asserts personality. The New Yorker has a writing style guide, fact-checkers, formatting rules. But what makes it feel like The New Yorker is the editor's taste applied consistently across hundreds of decisions. The style guide is necessary but not sufficient.
Failure here isn't a broken build. It's a product where every individual change passed tests but the accumulation feels assembled rather than designed. Nobody pushed a bad commit. The voice just drifted. The interaction patterns became inconsistent. The density shifted without anyone noticing.
None of this is a narrowing of the designer's role. It's an expansion. The execution layer compresses and what opens up above it is the whole product. Specs, policies, taste, knowing when to slow down and say "we're building the wrong thing"... it's all product thinking applied to a new context.
The designers who come out of this well won't just be the ones who only picked up engineering skills through AI. They'll be the ones who developed product skills and instincts too. Not the full stack in the technical sense. The full stack for product development.
Now, this doesn't necessarily mean PM, designer, and engineer will collapse into one role, but the boundaries will become porous. Fluency across disciplines will become the norm. The people who thrive will be deep experts in one yet genuinely sharp across the others.
This shift won't happen everywhere at once. Some companies are already there. Others will take years. But the people who do well will be the ones building and shipping now. It's the fastest way to develop the instincts you'll need when the building is done for you. Learn the current, not the fish
