Justin Michael argues that replacing engineers with generative AI can erode the knowledge and ownership that keep fragile systems running. His experience with banking transfers supplies the example: people understand the flaws, notice exceptions and keep things moving.
- Managers can mistake working code and polished documents for all the expertise that produced them.
- Michael argues that layoffs remove system knowledge, while compulsory LLM supervision can distance remaining workers from their systems.
- He treats learning through building and responsibility for failures as part of the work, even when neither appears in output counts.
The concern is about retaining people who understand what they operate. Caveat: Michael provides no incident investigations linking the outages he lists to AI adoption; that causal connection remains speculation.
The 140-comment thread on Hacker News adds pushback on causation and a possible countermeasure.
What the thread adds
- george_max — “We’ve had spectacular infrastructure outages for decades; I think that a cluster of incidents in one week doesn’t establish that AI is directly responsible for declining reliability.” They also offer a countermeasure: “AI can also help reduce this debt by asking it what you need help understanding.”
- plandis — “Without root cause analysis of these various failures I don’t see how one can conclude it’s due to uncaring, unintelligent people.” Their counterexample: “I’ve seen plenty of failures that are due to unforeseen issues that come up despite having caring, intelligent people owning those systems.”
- raynr — “I mislike this focus on “smart people”, it’s besides the point.” Their alternative framing: “The point is that capitalism only rewards giving a shit about acts that generate revenue.”
- donatj — describes replacing human quality-assurance testers with AI: “We have been using this AI QA system for about a year and a half and it’s literally never had anything to say about my work.” They contrast that silence with regular human pushback. This is one workplace account, without an independent measure of defects missed.
- tekacs — argues that AI lets people who care become “asymmetrically, wildly effective”. raphlinus replies: “But take a closer look what it does to systems, processes, and communities.” Their disagreement separates an individual’s increased capacity from the effects on the organization around them.
HN handles are pseudonymous, per-comment scores are unavailable, and ordering follows HN’s ranking; this is a slice of the thread, not a consensus.