Software development is becoming a commodity. Software engineering is the one in trouble.
Some time ago Coinbase cut 14% of staff and began testing one-person pods: a single engineer directing fleets of AI agents across engineering, design, and product. 40% of their code is already AI-generated, targeting 50%+ by October.
That's not an outlier. It's a leading indicator. Writing code is being automated. Not fully, but enough that shipping a feature no longer costs the headcount it did two years ago. The work didn't disappear. It moved up the stack.
The market splits in two. Big tech keeps the deep computer-science roles: platforms, foundation models, distributed systems, developer tooling. The substrates everyone else builds on. Agents can't design a new storage engine or model architecture from scratch yet. Everyone else, the product companies, scale-ups, and startups, converges on the Coinbase shape. Small, AI-amplified pods optimizing for latency from idea to revenue.
Three roles dominate that second tier.
Product Engineer
Owns the product surface. Less time writing code, more time deciding what to build and verifying the generated implementation actually solves the problem.
Quality Engineer
Not legacy QA. The human validating machine output at scale: evals, observability, security, performance. When most code is generated, "does it actually work" becomes the bottleneck.
Commercial Engineer
Builds the commercial machinery: pipelines, attribution, outbound automation, ad-tech, the tooling that scales acquisition. Not the marketer, but the engineer who builds what they run on.
What this means?
The middle hollows first. Pure-execution roles, implement the spec but don't shape it, compress first. Value concentrates at the edges: deep technical work agents can't touch, or product/quality/commercial work where judgment compounds.
You don't have to pick today. You do have to know which edge you're moving toward.
It won't arrive on a date and no one will announce it. One reorg, one hiring freeze, one "we won't backfill that" at a time. Pick your edge before you need it.