Stop Asking If AI Replaces Programmers. Start Asking Which Ones.

Every week we talk to engineers who are quietly terrified and engineers who have never been more in demand. Same industry, sometimes the same companies, completely different experiences of the same market. The ones struggling are engineers whose value was always about execution speed: writing code faster than the next person, implementing specs cleanly, shipping volume. The ones thriving built something different over their careers, the judgment to know what to build, whether it works, and what breaks under real production conditions.

AI made the difference between those two things impossible to ignore. The conversation in tech keeps circling the same question: will AI replace programmers? The question produces useless answers because it treats programmers as one thing and AI as one force acting on all of them uniformly. The market has already split along a specific fault line. Identifying which side of it you sit on is the more useful question.

The Numbers Look Contradictory. The Explanation Is Simple.

Software engineer job listings jumped 30% in 2026, with more than 67,000 open roles tracked across major tech companies, the strongest demand in over three years. At the same time, 52,000 tech workers lost jobs in Q1, with nearly half of those cuts attributed directly to AI. Both figures are accurate. They describe two completely different groups of engineers.

The roles getting cut skew heavily toward implementation work: junior developers on well-specified tasks, generalist mid-level engineers whose primary output was code volume, QA built around manual testing. The 67,000 open roles with thin candidate pipelines skew toward judgment work: cloud infrastructure, security engineering, AI integration, systems architecture, and engineers who can govern what AI produces rather than just produce alongside it.

The Bureau of Labor Statistics projects 17% growth for software developers through 2033, adding more than 327,000 jobs in the US alone. The profession is splitting, and the two halves are moving in opposite directions simultaneously.

What AI Has Actually Taken Over and What It Consistently Gets Wrong

The fault line becomes concrete when you look at what AI does reliably in production environments versus where it still fails in ways that cost teams real money. We see both sides of this every week across the teams we place engineers into.

AI handles these tasks well in 2026:

• Writing boilerplate endpoints, scaffolding components, generating unit tests from existing functions
• Translating code between languages and explaining unfamiliar codebases to new team members
• Implementing well-scoped, bounded tasks where requirements are explicit and success criteria are measurable
• Autocompleting repetitive patterns and drafting documentation from existing code

Where AI still fails consistently:

• System architecture decisions where constraints are implicit and tradeoffs are organizational rather than technical
• Debugging concurrency or distributed systems issues that surface only under real production load
• Security threat modeling specific to a given stack, attack surface, and compliance environment
• Translating what a stakeholder said they wanted into what the business actually needs, when those two things differ
• Owning the outcome when something breaks at 2am and the answer exists nowhere in the documentation

Engineers whose work lived primarily in the first list faced automation pressure long before AI coding tools existed. Engineers whose work lives in the second list are in the strongest hiring position of the past decade, because every team deploying agents needs more judgment sitting above the automation.

Why AI Growth Keeps Expanding the Demand for Engineers

The instinctive read of AI productivity gains is that companies need fewer engineers to produce the same output. Economic history consistently shows the opposite. The Jevons Paradox describes what happens when a resource becomes cheaper and faster to use: total consumption of that resource increases rather than decreases.

Applied to software development: when AI makes building software ten times faster and cheaper, companies build ten times more software. Every business that previously found the engineering cost of a custom internal tool, a client-facing product, or an automation pipeline now can afford to. The total demand for software is expanding faster than AI can compress the labor required to produce it.

Anthropic’s CEO stated in early 2026 that AI would handle most software engineering tasks within months. Anthropic simultaneously posted 448 open engineering roles. The work AI automates creates the conditions for more work that requires human judgment. The engineers who understand that dynamic are the ones with the clearest career path right now.

How the Split Shows Up When We Screen Candidates

We screen engineers for a living, and the fault line shows up in every batch of candidates. Seniority level and educational background rarely predict which side someone is on. The tell is in how candidates talk about the work they have done.

Engineers on the compressed side of the market describe their work at the output level. Features shipped, tickets closed, PRs merged. Ask them to explain an architectural decision and the answer gets thin fast: why that structure was chosen, what tradeoffs were made, what they would do differently with hindsight. A significant portion of their output over the past two years was AI-generated code that moved through a review process without anyone deeply owning the reasoning inside it. That gap is understandable. The workflow just never required accountability for understanding, only for delivery.

Engineers on the growing side of the market describe problems, failures, and decisions. A system that broke in production and what tracing the failure taught them. An architectural call they pushed back on and why. A constraint that forced a tradeoff and how they reasoned through it. Ask them to walk through a decision under pressure and they can, without referencing what a tool suggested. AI is a major part of their workflow, but the judgment layer sits entirely with them. That profile is what the market is paying a premium for and cannot source fast enough.

What to Do With This If You Are an Engineer Building a Career

Optimising for output volume in 2026 is optimising for the thing AI already does better than any human. The engineers who will look back at this period as a turning point are the ones who spent it building the judgment layer rather than the throughput layer.

In practice, that means owning what you ship end to end. When AI generates the code, read it carefully enough to debug it alone. The understanding is the professional asset. Accumulate production experience with systems that fail in ways worth learning from: incident response, architectural decisions under real constraints, debugging that requires reasoning from first principles rather than querying a tool. Those experiences build the pattern recognition that makes an engineer irreplaceable.

Move up the abstraction stack with intention. The engineers with the widest moat in this market are directing AI tools rather than competing with them. System design, security, domain expertise in specific industries: these sit above what AI can reliably automate, and the engineers who develop them early are the ones recruiters cannot find enough of.

Every engineer has a version of the same decision in front of them right now. Build the skills that sit above the automation line, or keep optimising for the skills that sit below it. The market will make the implications of that choice clear soon enough.

TechX trains and deploys engineers who have built the judgment layer alongside the technical one: people who can own production systems, govern AI output, and contribute from day one in the roles where demand is real and the candidate pool is thin. Whether you are hiring for those roles or building toward them, let’s talk.