In 1967, producing a Beatles album required a recording engineer, a mixing engineer, a producer, session musicians, an arranger, and a mastering specialist. Each role existed because the tools demanded it — operating a mixing console was a full-time skill, and mastering required specialized hardware that cost more than a house.
Today, Grammy-winning albums get produced in bedrooms. Not because the roles stopped mattering, but because the tools collapsed the barriers between them. One person with judgment and taste can now do what required a team of specialists.
Software teams are about to have their bedroom studio moment. Most of them just don’t know it yet, because the org chart was last updated when someone still thought “Agile transformation” was a strategy.
The Specialization Was Never About the Work
Product management, design, engineering, and DevOps didn’t become separate disciplines because the work is fundamentally different. They became separate because the tools were hard.
Learning Figma well enough to produce production-quality UI took months. Understanding Kubernetes took a year. Writing a proper PRD required knowing how to think about user problems in a structured way that most engineers never practiced. Each skill had a steep enough learning curve that specialization was the rational response.
AI flattens those curves.
A developer with Claude Code can now produce a polished UI that would have required a designer to refine. A product manager can prototype a working feature to validate an idea before engineering touches it. An engineer can configure a deployment pipeline by describing what they want rather than memorizing YAML syntax. The YAML syntax part alone should be worth a Nobel Peace Prize.
The multiclass penalty — borrowing from D&D — is disappearing. You used to pick Fighter and accept that you’d never cast Fireball. Now the AI is handing out free levels in every class. Your Paladin can write CSS. We live in unprecedented times.
What’s Actually Converging
This isn’t theoretical. Watch what’s already happening.
Product + Engineering. Product managers who can prototype are testing ideas in hours instead of waiting for sprint planning. The feedback loop collapses from weeks to minutes. Engineers who understand user problems make better architectural decisions because they’re solving the problem directly — no more translating requirements through three layers of Jira tickets like the world’s least fun game of telephone.
Design + Engineering. The gap between “mockup” and “working code” is vanishing. AI generates a functional component from a design description faster than a designer can create the mockup for an engineer to misinterpret. The translation layer — where intent went to die — disappears.
Engineering + DevOps. Infrastructure-as-code was step one. AI is step two. Developers who can describe their deployment needs in plain language don’t need a dedicated platform team to write their Terraform. Somewhere, a DevOps engineer just felt a disturbance in the Force.
But Convergence Without Taste Is Just Chaos
Here’s the part most people skip past. AI lowers the floor for executing across domains. It does not raise the ceiling for judgment within them.
A developer can now generate a landing page in minutes. Whether that page communicates the right value proposition, guides the eye correctly, and feels trustworthy — that’s taste. AI didn’t provide it. The developer either has it or they don’t. And if they don’t, they just shipped a technically functional page that makes your product look like a 2009 WordPress template. Fast.
The same applies in every direction. An engineer can prototype a feature, but knowing whether that feature solves a real problem or just a hypothetical one is product judgment. A product person can ship working code, but knowing whether the architecture will survive six months of iteration is engineering judgment. AI lets you do things quickly. Taste is what keeps those things from being terrible.
The risk of convergence without taste is people doing more things badly. A mediocre landing page, a fragile architecture, a deployment pipeline held together with hope — all shipped by one person who had access to every tool and none of the instincts. This is the Saruman problem: immense power, zero wisdom, and eventually the Ents show up.
Taste is knowing what good looks like across the full surface. It’s the designer’s eye for hierarchy, the engineer’s instinct for maintainability, the product thinker’s discipline to say no, the operator’s refusal to ship something that can’t be monitored. Those instincts don’t come from AI. They come from experience, from caring, and from paying attention to things most people consider someone else’s problem.
The people who thrive in this convergence won’t be the ones who can do the most things. They’ll be the ones with the best judgment about which things to do, and what “done well” looks like in each domain.
The New Archetype
The role that emerges isn’t a generalist. Generalists knew a little about everything and nothing deeply — the party member who put one point into every skill and wonders why they can’t pick any lock or persuade any guard. This is different.
I’d call it a product engineer: someone with strong technical fundamentals, taste across multiple domains, and enough breadth to operate across the entire delivery chain. They understand user problems. They can evaluate whether a solution looks and feels right. They can build it. They can ship it. And critically — they know when something isn’t good enough, even when it technically works.
Netflix calls a version of this the “full-cycle developer.” Spotify had “autonomous squads.” Those were organizational experiments bolted onto the old specialization model. AI makes the underlying capability shift real.
What This Means for Teams and Hiring
Teams get smaller. The five-person cross-functional squad — PM, designer, two engineers, DevOps support — becomes two or three product engineers who cover the full surface. Not because you’re cutting corners, but because the coordination overhead of handoffs between specialists was always the actual bottleneck. The meetings about the meetings were not, it turns out, load-bearing.
Brooks’s Law says adding people to a late project makes it later. The corollary nobody talks about: removing handoffs from an on-time project makes it faster.
For hiring, stop filtering by role labels. Start filtering for taste and learning velocity. Ask candidates what good software feels like. Ask them about a decision they’d make differently with hindsight. Ask what they’d refuse to ship. If they look at you like nobody’s ever asked that before — that tells you something.
The engineer who says “that’s not my job” when asked about user experience is going to struggle. The designer who can’t read a pull request is going to struggle. Look for people who are curious about the whole problem — and who have opinions about what quality means across all of it.
In Summary
Specialization in software was a response to tool complexity, not to the nature of the work itself. AI is removing that complexity. But the thing that makes convergence work isn’t the AI — it’s the taste and judgment of the person wielding it. AI gives you the power. Taste tells you what to do with it. The roles aren’t disappearing. The walls between them are. The people who thrive will be the ones with the best instincts on both sides of every wall.