Adrian Quiroga found Codex-generated code failing an automated quality check. After inspecting it, he judged the behavior correct and asked Codex to refactor it into ten files. The encounter with unfamiliar code frames his question: what must engineers understand when agents write faster than people can read?

His answer has three parts:

  • Fundamentals: networking, concurrency, and algorithms help you recognize when a system behaves strangely.
  • Breadth: a map of the components and their relationships tells you where to investigate.
  • Just-in-time depth: learn an unfamiliar subsystem deeply when the problem requires it, and recognize when you need help.

Quiroga’s proposed owner can direct an investigation without having personally written every module. His claim that companies will stop funding the old apprenticeship is a forecast, not evidence from a hiring study.

The 86-comment thread on Hacker News adds concrete review practices and challenges the training story.

What the thread adds

  • CGamesPlay — a rule for choosing what to inspect: “I review all changes that touch persisted data formats, wire protocol, and/or public API surface”. They describe these as boundaries where mistakes have expensive consequences.
  • mrothroc — a deliverable that makes the system map tangible: “I’ve encouraged developers on my team to make sequence diagrams part of their deliverable”. These diagrams show how components exchange messages.
  • smallerfish — a recurring maintenance practice: “On AI heavy projects I tend to pause feature development every few days for code & schema review, and subsequent cleanups.”
  • trjordan — dissent about what counts as failure: “The failure that actually matters here is product failure. Angry users. Security incidents. Downtime. Broken workflows.” They report knowing companies with stressed developers, faster shipping, and fewer incidents; that remains anecdotal.

Who learns to investigate?

Two commenters raise the same training gap. scottLobster asks, “And how will you have the skill to go hunting inside the box if all you ever do is high level systems work?” ArcHound asks, “How do you maintain your expertise when you stop doing technical work?” Quiroga names the capabilities an owner needs, but leaves their long-term development unresolved.

HN handles are pseudonymous; no per-comment scores are public. Ordering is HN’s own ranking. This is a slice of the thread, not a consensus.