AI Matches Median Engineers on Code by Late 2025, Upending Software Economics

3 articles · Updated · O'Reilly Media · Jul 27

By November 2025, AI code generation had reached roughly the quality of a median software engineer for common patterns, while producing code far faster and at much lower cost.
Opus 4.5 marked the tipping point, but the shift had been building through 2025 as agentic harnesses, tool use and function-calling made AI coding broadly usable.
That collapse in code-production costs turns code from a scarce asset into a disposable, regenerable artifact, the author argues, pushing teams to focus less on handcrafting code and more on validation.
Production testing, observability, behavioral checks and other evaluation systems become more important in that model, because nondeterministic AI output demands stronger engineering discipline rather than less.
For 2026, the piece argues the industry’s advantage will shift to teams with fast feedback loops and durable operational rigor, not simply those generating the most code.