In 1933, German General Kurt von Hammerstein-Equord published a framework for categorizing military officers. Two axes: clever or stupid, lazy or industrious. Four quadrants, each with a clear prescription.
— Kurt von Hammerstein-Equord, 1933Those who are clever and industrious I appoint to the General Staff. Use can under certain circumstances be made of those who are stupid and lazy. The man who is clever and lazy qualifies for the highest leadership posts. He has the requisite nerves and the mental clarity for difficult decisions. One must beware of anyone who is stupid and industrious - he must not be entrusted with any responsibility because he will always cause only mischief.
The clever-and-lazy commander doesn’t waste energy on busywork. He finds efficient solutions because doing things the hard way is too much effort. The stupid-and-lazy officer is harmless: limited capability, limited action. But the stupid-and-industrious officer is the most dangerous person in any organization. He doesn’t know what he’s doing, and he does it with relentless energy.
This framework is 93 years old. It has never been more relevant.
The Matrix
The interesting thing about AI agents is that they don’t change which quadrant you’re in. They amplify it. Google’s DORA 2025 research found that AI’s impact depends less on the tools and more on the engineering culture deploying them. Mature teams see AI multiply their effectiveness. Immature teams see it multiply their dysfunction.
AI is a force multiplier. A force multiplier on zero is still zero.
Clever + Lazy: The Commanders
Larry Wall listed laziness as a core programmer virtue: “the quality that makes you go to great effort to reduce overall energy expenditure.” Bill Gates supposedly said he’d choose a lazy person for a hard job because they’d find the easy way. Hammerstein-Equord called them fit for the highest command.
In March 2026, the clever-and-lazy engineer writes specs, not code. They build AGENTS.md files and completion criteria. They design the constraints that make agents reliable rather than babysitting every keystroke. OpenAI’s Codex team built over a million lines of production code where zero lines were written by human hands. The engineers designed the environment. The agents wrote the code.
Kent Beck’s framework captures why this works. AI deprecates syntax mastery, API memorization, and boilerplate. It amplifies vision, architectural strategy, taste, and judgment. When anyone can build anything, knowing what’s worth building becomes the skill.
— Kent BeckIn a world of abundant cheap code, what becomes scarce? Understanding. Judgment. The ability to see how pieces fit together.
The clever-and-lazy engineer doesn’t produce less. They produce leverage. They spend their energy on the 20% of decisions that determine whether the other 80% goes well or badly. Then they let an agent handle the 80%.
Clever + Industrious: The Staff Officers
The capable engineer who ships a lot of high-quality code. With AI, they ship even more. They build elaborate CI pipelines, exhaustive test suites, microservice architectures. Everything is well-crafted. Everything is more than what was needed.
Hammerstein assigned these officers to the General Staff: useful, productive, necessary. But not command. The clever-and-industrious engineer’s failure mode is over-engineering. They plan for 100 scenarios; the business invents the 101st. The sense of control is an illusion.
With AI agents, this archetype can now over-engineer at 10x speed. The Cortex 2026 benchmark tells the story: PRs per author up 20% year-over-year, but incidents per PR up 23.5% and change failure rates up 30%. More output. More breakage. The clever-and-industrious engineer is producing higher-quality pieces that don’t fit together into a simpler whole.
The difference between this quadrant and the commanders: the commander knows when to stop. The staff officer doesn’t.
Stupid + Lazy: The Harmless
The engineer who doesn’t use AI, doesn’t learn new tools, coasts on existing skills. Stack Overflow’s 2025 survey found that while 80% of developers use AI tools, trust in AI accuracy has fallen from 40% to 29%. Senior developers are the biggest holdouts, and their resistance creates cultural signals that AI adoption is optional.
Hammerstein said these officers were suited for routine duties. He wasn’t wrong. They won’t break anything. They also won’t build anything new. In an era where AI is widening the gap between those who leverage it and those who don’t, this quadrant is slowly becoming untenable. But at least they’re not causing mischief.
Stupid + Industrious: The Dangerous
This is the quadrant Hammerstein warned about. Remove them immediately. Don’t reassign them. Don’t try to train them. Remove them.
In March 2026, the stupid-and-industrious engineer vibe-codes everything. They commit AI-generated code without reading it. They ship 50 PRs a day that all need to be reverted. They use AI to look productive without being productive.
The data is brutal:
- CodeRabbit (Dec 2025): AI-generated PRs contain 1.7x more issues, 2.74x more XSS vulnerabilities, 3x more readability problems
- GitClear: For the first time in history, developers paste code more often than they refactor or reuse it. 4x more code cloning.
- Veracode: 45% of AI-generated code contains OWASP top 10 vulnerabilities
- 18 CTOs surveyed (Aug 2025): 16 reported production disasters directly caused by AI-generated code
In January 2026, a platform called Moltbook launched with an AI-built codebase. Three days later it had leaked 1.5 million API keys. The AI agents generated functional database schemas but never enabled Row Level Security. No human verified it.
In July 2025, an autonomous Replit agent at SaaStr executed DROP DATABASE on production during a code freeze. Then it generated 4,000 fake user accounts and fabricated logs to cover its tracks. The agent’s own explanation: “I panicked instead of thinking.”
It can’t think. But the engineer who deployed it without guardrails couldn’t either.
METR’s controlled trial found AI made experienced developers 19% slower. Before the study, they predicted a 24% speedup. After the study, having seen their own timing data, they still believed AI had made them 20% faster. A 39-percentage-point gap between perception and reality. The stupid-and-industrious engineer doesn’t know they’re causing mischief. They think they’re the most productive person on the team.
The Widening Gap
The stupid-and-industrious engineer has always existed. What’s new is the blast radius.
Before AI, a bad engineer could produce maybe 2-3 bad PRs a day. The damage was rate-limited by typing speed. Now they can produce 50. The joke that “two engineers can now create the tech debt of fifty” contains more truth than humor.
Open source is already drowning. Daniel Stenberg shut down cURL’s bug bounty after AI-generated spam hit 20% of submissions. Mitchell Hashimoto banned AI code from Ghostty. Steve Ruiz closed all external PRs to tldraw. Craig McLuckie observed that “good first issue” labels now get “inundated with low quality vibe-coded slop that takes time away from doing real work.”
Meanwhile, Karat’s research shows AI is widening the engineering skills gap faster than education can adapt. Employment among software developers aged 22-25 fell nearly 20% between 2022 and 2025. The juniors who would have learned by doing are being replaced by agents. The seniors who know what to build are more valuable than ever.
— Stack Overflow Blog, December 2025No developer will become a senior engineer by skipping their junior development years because genAI did all the work.
AI writes code optimized for the happy path. Senior engineers write code optimized for the moment the happy path ends. That’s the gap, and AI is making it wider.
Hammerstein’s Prescription
The general was clear about the stupid-and-industrious officer: remove them immediately. But the modern version of this isn’t firing people. It’s about which quadrant you choose to operate in.
The same engineer can be clever-and-lazy or stupid-and-industrious depending on how they work. Write a spec before prompting, review every diff, set explicit constraints: you’re in the commander quadrant. Skip permissions because it’s faster, commit without reading, let the agent “just handle it”: you’ve moved to the dangerous quadrant.
The tool doesn’t determine the quadrant. You do.
Hammerstein’s clever-and-lazy commander had “the requisite nerves and the mental clarity for difficult decisions.” The difficult decision in March 2026 isn’t which model to use. It’s knowing when the right output is no output at all.
Related Posts
