At QECon, a veteran backend speaker threw out six engineering practices being repriced by AI: the faster AI writes code, the more those old-school fundamentals shift from nice-to-have to survival line.

What this is

This is a counterintuitive phenomenon. The narrative shouts that AI writes code and engineers lose jobs, but this speaker offers the opposite observation: after AI causes code output to explode, the engineering fundamentals long neglected have become the key to whether systems actually run.

  • Input validation: AI Agents (AI assistants that can independently invoke tools) pass weird boundary values and even hallucinated data. Interfaces must defend against AI the way they defend against hackers.
  • Logging and distributed tracing (end-to-end chain logging): AI fixing bugs can't see standardized chains and falls into error loops.
  • API documentation: Agents decide when to call based on field descriptions. Bad docs means handing the system to AI to click randomly.
  • Idempotency (same request executed multiple times yields consistent results): AI retries 100x more frequently than humans. Without it done properly, duplicate orders flash in an instant.
  • Domain-Driven Design (DDD, organizing code by business boundaries): AI context windows (how much code it can see at once) are limited. Spaghetti code is a nightmare for it.
  • Automated verification (Harness engineering, letting AI run tests repeatedly in controlled environments): Cursor and Claude Code spit thousands of lines overnight. Machines must stand in for humans as the first line of defense.

Industry view

The speaker takes neither the "AI replaces engineers" nor "AI is useless" position, but a third path—AI amplifies the value of engineering fundamentals. This aligns with what Andrew Ng and others have repeatedly emphasized this year: Agent projects get stuck at the final step of deployment, usually not because of model capability, but insufficient engineering capability.

The counter-voice is worth hearing. A cloud vendor infrastructure engineer says this framework holds for top-tier teams but is a luxury for most mid-sized and small companies—they never properly patched logging and validation in the first place, AI arrives and they have even less motivation, and short-term they'll experience the chaos of bad code combined with AI bad output. Another underestimated risk: once basic skills are outsourced to AI training and toolchains, will the next generation of engineers' core competencies regress? The article doesn't answer that question.

Impact on regular people

  • For enterprise IT: Engineering debt postponed over the past 5 years will be actively demanded by business departments to be repaid, otherwise AI tools can't connect.
  • For individual careers: Pure business code-writing roles are under pressure; engineers who understand contracts and can feed systems to be AI-readable become scarcer.
  • For consumer market: Short-term users will perceive more bugs and duplicate charges—when AI auto-executes tasks, if the backend hasn't done idempotency well, it's consumers' money and orders that get hurt.