AI Is Splitting Development Teams: The Joyful “Lazy” vs the Broken “Craftsmen”

The Lazy’s Celebration

Deedy Das describes a new class of engineers he calls the “Lazy” (or “托管派”). Their workflow is entirely AI‑driven: they glance at a Jira ticket, hand the description to Claude, generate code with a single click, and submit a pull request without testing, refactoring, or even reviewing the changes. When senior engineers raise concerns, they simply forward the question to the AI, paste the answer back, or let the AI draft a polished stand‑up report. In their view, software engineering has become a performance to prove they are “smart and hardworking” by offloading everything to AI.

The Craftsmen’s Collapse

The opposite group, the “Craftsmen” (or “守夜人”), bears the cognitive overload of maintaining and debugging the AI‑generated code. Their PR queues contain dozens of reviews, and they spend hours dissecting hidden bugs in 20,000‑line PRs produced by the Lazy. After they propose thorough fixes, the Lazy discard the advice, ask the AI to rewrite the code, and merge the new, still‑buggy version. This creates a vicious loop where massive, low‑quality code (“Slop Cannons”) repeatedly floods the codebase, leading to system instability and eventual collapse.

The Cost of Tokenmaxing

Management promotes “Tokenmaxing” – the belief that more tokens (AI usage) and higher LOC growth equate to higher productivity. They even deploy AI‑powered code‑review bots to evaluate AI‑written code, creating a feedback loop where models audit themselves. In large, bureaucratic firms, this metric drives impressive PPT charts of doubled commit volumes, while months later the system crashes under the weight of hundreds of thousands of lines of opaque, “black‑box” code.

The Vanishing Mentorship

Traditional mentorship—code reviews, design discussions, and painful debugging—has eroded. Junior developers no longer seek guidance from seniors; they request instant patches from Claude without understanding the underlying mechanics. This accelerates skill atrophy: after a few years, engineers who rose from junior to senior can only write prompts, lacking the ability to diagnose network issues, deadlocks, or memory leaks.

Conclusion: Reclaim Craftsmanship

Das warns that without a conscious effort to restore craftsmanship—insisting on reading, testing, and understanding every change, rejecting blind AI‑generated code, and preserving human judgment—software development risks becoming a “technical hell” of black‑box code. The article calls for engineers to reclaim ownership of their code and rebuild the spirit of craftsmanship.