The great AI delusion is falling apart

Individual productivity vs personal incentives

• Many argue there’s little personal benefit to being more productive with AI if pay and time off stay the same; faster workers just get more tasks, not more reward.
• Others say the “reward” is job security when peers are less productive, but this is seen by some as a poor consolation.
• Some claim to use AI to dramatically increase throughput (more commits, more releases, multi-backend migrations) and are unbothered by systemic issues; others question whether those metrics reflect true value or quality.

Perceived vs actual productivity (METR study)

• The METR RCT showing developers felt ~20% faster with AI but were ~19% slower is heavily debated.
• One camp sees it as overdue counterweight to hype and proof that self‑reported productivity is unreliable.
• Critics note small sample size, unfamiliarity with tools, and argue it mainly shows how hard it is to measure software productivity.
• A key concern: developers are poor judges of whether AI use is speeding them up, so “it feels faster” isn’t evidence.

Experiences with AI coding tools

• Positive reports: building complex systems one wouldn’t attempt otherwise; faster boilerplate and mundane tasks; reduced cognitive load even if wall-clock speed is unclear.
• Negative reports: frequent low‑quality suggestions, extra time validating/fixing output, atrophy fears, and a “slot machine” effect where occasional wins mask overall slowdown.
• Some think modest, task‑specific gains (autocomplete++ for simple work) are undeniable; large, general productivity leaps are disputed.

Interviews, skills, and tool use

• Multiple commenters now allow LLMs in coding interviews and observe candidates hurting themselves: fumbling with tools, pasting bad code, failing to reason about it.
• These interviews are used to distinguish people who can truly code and critically evaluate AI output from those who can only prompt and copy‑paste.
• Consensus: even if AI is “the future”, humans still need deep understanding to review, debug, and adapt generated code.

Economic, societal, and hype dynamics

• Some see AI as underhyped, akin to a tectonic shift; others compare its hype cycle to Segways, NFTs, or overblown “web3” claims.
• Concerns include: capital misallocation and potential economic shocks; noise and “AI slop” (spam emails, bogus legal threats, resumes) degrading everyone’s productivity; environmental and labor impacts.
• There’s tension between workers who logically fear job‑threatening productivity gains and those who view AI as analogous to open source: expanding what’s possible and ultimately increasing demand for software work.