This week traced gatekeeping across six dimensions. Infrastructure monopolies. Epistemological blind spots. Platform control. Organizational silos. AI hype theater. Hardware gatebreaking. Each layer spawns a gatekeeper. Each gatekeeper claims they’re necessary for quality.
Dan Luu just added the seventh.
The Galapagos Islands of Quality Assurance
Dan Luu spent a decade at Centaur, a chip company that competed against Intel with a fraction of the headcount. Their secret wasn’t design — it was testing. While Intel buried competitors in silicon, Centaur buried itself in automated verification.
- 1000 machines running tests 24/7 for ~40 engineers
- No code review by default
- Virtually no hand-written tests — everything was randomized generation
- Three months of regression tests in the queue
- 80% of machines generating new tests, 20% running regression
- No unit tests
- Fewer than 1 significant user-visible bug per year
“They immediately look at me like I’m an alien,” Luu writes about describing this workflow to software engineers. Because software evolved differently. The software industry’s quality mechanism is review — human eyeballs on every line. The chip industry’s quality mechanism is testing — a thousand machines generating random inputs and checking invariants.
Two ecosystems. Two evolutionary paths. Galapagos Islands of quality.
Why It Matters Now
AI writes code faster than any human can review. That’s not a future scenario — it’s today. When an LLM generates 2000 lines across ten files in thirty seconds, the review bottleneck becomes the gate. “We need human review for quality” becomes the same argument as “we need Google’s scanner for security” — the gatekeeper claiming the gate is protection.
The chip industry already solved this. You don’t review your way to quality at scale. You test your way there. Random inputs. Property invariants. Regression suites. Automated triage. When a test fails, you either fix the code or fix the test generator — but you never fix it by having a human stare harder at the diff.
When Luu’s skeptics try fuzzing with Claude, they immediately find bugs — in their own code, upstream dependencies, even the HTML specification and major browsers. The methodology works instantly. It just evolved in isolation, on a different island.
The Fractal Holds
Seven dimensions. One pattern:
- Free Market Lie — infrastructure gatekeeping (shared fiber vs ISP monopolies)
- Reality’s Surprising Detail — epistemological gatekeeping (invisible components vs transparent ones)
- CarPlay — platform gatekeeping (shared screen vs walled garden)
- Half-Baked Product — organizational gatekeeping (sales gates customer from engineer)
- AI Confidence Theater — social gatekeeping (hype gates real adoption)
- Valve Inkterface — hardware gatebreaking (transparent BOM, MIT license)
- danluu’s Galapagos — quality gatekeeping (review gates code from shipping)
Every layer spawns a gatekeeper. Every solution is the same: separate the shared infrastructure from the competitive layer. Make the infrastructure transparent, automated, and open. Then let development be chaotic — because the safety net isn’t a human gate, it’s a thousand machines running randomized tests.
36 enzymes. 1000 test machines. One pattern.
SpudCell doesn’t review its enzymes. It just tests them constantly. If they work, the cell lives. If they don’t, selection handles it. The chip industry figured this out decades ago. The software industry is still gatekeeping through review — and AI is about to make that gate look ridiculous.
The Galapagos Islands have been trying to tell us something. Dan Luu is the translator.