A year after the 100% AI mandate nobody understands our code
A Reddit post claiming that a company mandated 100% AI-generated code prompts debate over whether aggressive LLM adoption is destroying maintainability, eroding developers’ skills, and turning codebases into opaque “AI slop.” Commenters weigh the trade-off between short-term speed and long-term technical debt, arguing over when (or if) it’s wise to go “all in” on agents, and whether management is driven more by FOMO and metrics than product quality or employee growth. Some also question the authenticity of such viral horror stories, noting the rise of AI-generated anti-AI content and unreliable detection tools, while others insist the underlying problems—loss of understanding, weak design, and misaligned incentives—are very real.
AI Adoption Timing and Strategy
- Many argue “going all‑in” on AI a year ago was risky given rapid model deflation: the same spend buys much more capability later.
- Others counter that waiting forever is also irrational: time-to-market, competitive pressure, and learning by doing all matter.
- Some advocate letting others “Leeroy Jenkins” first, then copying best practices; others reply you can’t know when maturity has arrived and late adopters will still repeat old mistakes.
- Comparisons are made to Moore’s law: for most software uses today, it’s better to start now than to wait for cheaper tokens.
Code Quality, Understanding, and Tech Debt
- Strong theme: AI accelerates output and tech debt. Organizations pile up huge, incoherent codebases where no one understands the architecture.
- Several report “vibe‑coded” AI-generated systems: large, messy PRs, feature rewrites, duplication, and brittleness where only an LLM can feasibly debug or modify code.
- Others note that “nobody understands the whole codebase” was already common in large systems pre‑AI; AI mainly scales an existing problem.
Developer Experience, Skills, and Incentives
- Multiple comments describe AI mandates making programming joyless: long days spent prompting, waiting, and reviewing instead of learning or designing.
- Concern that reliance on AI erodes individual skill, cognitive engagement, and career differentiation. Engineers are seen as being used to optimize company metrics (PR count, feature velocity) at the expense of craftsmanship.
- Some say individuals can still use freed time to deepen understanding and let AI handle “boring” bits; others respond that management metrics and mandates make that unrealistic.
How AI Should Be Used
- Several advocate using LLMs as assistants for analysis, refactoring, documentation, and small tasks, not as autonomous agents owning entire codebases.
- Suggestions include literate programming and using agents to generate high-quality documentation, then refactoring iteratively.
Authenticity and “AI-Generated” Discourse
- A substantial subthread debates whether the original Reddit post itself was AI-generated, citing an AI-detection tool and stylistic “tells.”
- Others question detector reliability and worry that labeling disliked content as “AI” becomes a lazy dismissal tactic.
- Broader unease emerges about AI-written engagement bait, the “dead internet” feeling, and humans adopting LLM-like writing styles.