The "Learn to Code" Era Is Over — Here's What Replaced It
For fifteen years, the advice was: get a CS degree, learn React, ride the bull. The bull is gone. The new advice is unsexy, harder to market, and far more honest about what actually pays.
Between roughly 2010 and 2022, software engineering was the single best risk-adjusted career bet a normal person could make. Bootcamp graduates with no degree were clearing $120k. Mid-career engineers at FAANGs were clearing $400k. Entire pundit careers were built on telling laid-off journalists to learn to code as if it were a moral cudgel rather than a market timing observation.
That market is gone. Not because programming died — programming is more powerful than ever — but because the part of programming that paid extraordinarily well for being a normal-talent practitioner has been compressed by the same force that made every previous golden trade boring: too many entrants, automation of the average case, and the migration of value to a smaller set of people doing harder things.
What changed
Three things, all of which compounded.
1. Supply caught up. A decade of "learn to code" produced more new bootcamp graduates per year than the entire CS-degree pipeline of the 1990s and 2000s combined. The average application got worse, the average resume got more samey, and the floor for getting a first job rose accordingly.
2. Capital tightened. Zero-interest-rate VC funded entire categories of products that didn't need to make money. Those categories funded the salaries that made even mediocre engineers wealthy. When rates rose, the categories died, the salaries normalized, and the marginal engineer became actually marginal again.
3. AI compressed the easy half. The kind of code most junior engineers were paid to write — CRUD handlers, glue logic, form validation, undifferentiated React — is precisely the kind of code that LLMs produce competently. Not perfectly. Competently, which is the same thing the median junior was producing, just faster and cheaper. The market noticed.
What replaced it
The new economy still rewards technical work, but it rewards it in a different shape. Roughly:
- Taste, judgment, and integration — the engineer who can look at a half-broken system, decide what's worth fixing, ignore the wrong distractions, and ship the right diff. AI can write the diff. It cannot decide which diff to write.
- Domain depth — the engineer who actually understands ad markets, supply chains, claims processing, FX settlement, or biology. The pure full-stack generalist is fungible. The engineer who can read a clinical-trial protocol is not.
- Distribution — anyone who can ship product, attract users, and keep them. This was always undervalued by engineers and overvalued by markets, and the gap has only widened.
- The very, very hard problems — distributed systems at scale, kernels, compilers, ML systems, security at depth. These were always the high tier. They remain the high tier. They've been somewhat insulated from compression because they require real expertise that takes years to build.
What's not on the list: being a competent React developer, being a competent backend generalist, being a competent SRE who runs a runbook. Those jobs still exist. The wages are reverting to the boring mean.
The obvious counter
"AI will get worse at coding before it gets better, models hallucinate, the productivity gains are overstated." Maybe — but the question isn't whether AI is perfect; the question is whether AI is good enough to compress the median junior task, and the answer is observably yes. Companies aren't replacing senior engineers with AI. They are quietly hiring fewer juniors. That's the entire mechanism.
The danger to the median engineer is not that AI replaces them — it's that AI replaces the job they were going to be promoted into next year.
"Software is eating the world; demand for engineers grows forever." Demand for software grows. Demand for the specific job called "software engineer" is downstream of that. The 1900s were the age of electricity, and electrical engineering became important — but building wiring stopped being a path to wealth pretty quickly once the supply caught up. We are at the wiring stage of a lot of programming.
The response
If you're entering the field today, the honest framing is:
- Learn to code anyway — but as literacy, not as a career floor. It's the ability to glue your career together regardless of what you actually do.
- Pick a domain. Become annoyingly deep in something the market pays for. Insurance, biotech, energy, defense, logistics. The boring industries are where the unautomated value lives.
- Build something. The portfolio of "shipped a real product, got real users, kept it alive" is now worth more than three internships.
- Get good with AI tooling. Not as a parlor trick — as the new baseline tool of the trade. The engineers who refuse to use it are about to look the way the engineers who refused to use IDEs looked in 2005.
The "learn to code" era was a trade — a real, lucrative, time-limited trade — and people who caught it early did extraordinarily well. Trades end. The next one rewards judgment and depth, neither of which fits on a billboard, neither of which can be bootcamped in nine weeks, and both of which are how the next generation of well-paid technical careers will actually be built.