For years, I watched engineers pick one stack and go deep. Backend or frontend, one language, one framework, and they stuck with it. They became wizards in their corner of the system. And it worked: specialists climbed the career ladder faster than the generalists who spread themselves across languages and layers.
Then I noticed a slow shift. A handful of companies, mostly FANG, started rewarding generalists: engineers who could touch CI/CD, backend services, and frontend forms in the same week. The rest of the industry was slower to catch on.
That shift isn’t slow anymore. Generalists are winning.
The cost of staying narrow
Here’s the problem with deep specialization: if you’re locked into one stack, you rarely look up. You don’t dig into your company’s infrastructure or its architecture decisions, because your stack never required you to. That gap didn’t matter much when a human wrote every line of code. It matters a lot now.
Build your own agent factory
Every engineer is now running a small factory of AI agents, whether they planned to or not. The equation is simple: the more productive, context-efficient, and comprehensive your agents are, the more productive you are. And a more productive engineer earns, and saves, more for the company. Running that factory well means understanding more of the system than any single stack ever demanded. That’s the generalist’s strength.
Product sense stopped being a bonus
One thing hasn’t changed: engineers who deeply understand the product were always the strongest ones. That’s still true. What changed is where the bar sits. Understanding the product well enough to propose real improvements used to make you an A-player. Now it’s the baseline. AI can handle the manual coding. It can’t decide what’s worth building. Connecting business value to a technical decision is still, and only, a human job.
The job got bigger, not smaller
We used to talk about engineers handling complex problems like performance issues. That’s still on the list. Agentic engineering adds more to it, and all at once: Total Cost of Ownership, the harnesses that run your agents, and the automation that keeps their output maintainable.
So, yes, code generation is solved
But engineering isn’t. Setting direction, designing the system, and verifying the result: that’s still the fun part, and it’s still ours. It also demands more experience, not less.
Inspired by "The New SDLC With Vibe Coding"