Thoughts on slowing the fuck down

Perceived decline in software quality

• Many see more brittle, failure‑prone systems; outages feel more frequent and visible due to consolidation (AWS, GitHub, Cloudflare) and tighter integration.
• Others argue software has always been “crap”; what changed is faster deployment, more automation, and less slack between failures.

Root causes: speed, incentives, consolidation

• Cultural push for “move fast” and C‑level pressure to use AI is eroding developer autonomy and review rigor.
• Capitalism and shareholder value are seen as driving optimization toward minimum acceptable quality (“enshittification”) and little incentive for reliability outside regulated sectors.
• Large shared infrastructure magnifies impact of failures, but posters stress the real issue is under‑engineering and weak accountability.

DevOps, Andon cord, and process discipline

• Several invoke Toyota‑style “Andon cord”: anyone can stop the line to fix root causes, leading to higher long‑term quality.
• Google‑style SRE practices (error budgets, forced pause to fix) are cited as examples of needed “adult in the room”.

AI coding agents: benefits and risks

• Many use LLMs successfully for boilerplate, exploration, refactors, documentation, and tests; some say they now one‑shot trivial changes to prod for low‑risk sites.
• Others report AI‑heavy stacks producing glaring bugs, memory leaks, and architectural “booboos” that compound until the codebase becomes unmaintainable.
• Strong split on review: some insist every AI‑written line must be reviewed and tested; others say that’s unrealistic and rely on tests and monitoring instead.
• Concern that managers believe unreadable code is fine “because AI will handle it”, leading to loss of human understanding and long‑term risk.

Is software engineering really “engineering”?

• Large sub‑thread debates whether software is an engineering discipline or a craft: lack of licensing, weak liability, and tolerance for shipping broken systems vs. civil/aviation software with formal methods and strict regulation.
• Some argue “engineering” is about risk management over time; by that standard, most commercial software falls short.

Labour, jobs, and economics

• Some openly accelerate AI integration knowing it will cut jobs, wanting to exit “bullshit jobs”.
• Others worry about short‑term mass displacement, vendor lock‑in to AI providers, and future price hikes once organizations are dependent.

Tools, meta‑work, and best practices

• Recurrent theme: the last decade has produced layers of meta‑tools, frameworks, and abstractions, often worsening complexity rather than solving real user problems.
• Suggested mitigations: slow down enough to think, constrain agent scope, keep changes small and reviewable, capture intent in living docs, and measure success by solving problems, not lines of code or feature velocity.