필사 모드: How Successful Companies Go Blind — and What It Looks Like in Engineering

• Introduction — the cavefish and competence blindness
• The cause is suppression, not neglect
• The legacy system nobody is allowed to question
• Dashboards stay green while reality rots
• When process becomes scar tissue
• Closing — how sight returns
• References

Introduction — the cavefish and competence blindness

Ian Reppel's essay "How Successful Companies Go Blind," which recently trended on Hacker News, opens with the Mexican cavefish (Astyanax mexicanus). The same species exists in two forms only kilometres apart: in the rivers of the Sierra del Abra it has eyes and behaves like an ordinary fish; in the limestone caves under the same mountains it is blind, depigmented, and translucent.

The important detail is how the eyes are lost. It is not passive disuse. Within hours of fertilization in cave conditions, the lens-building programme triggers early apoptosis — programmed cell death — and the energy that would have built optic tissue is redirected to traits the cave actually rewards: better smell, deeper feeding, and fat reserves for the next lean year. The eye is not neglected. It is actively switched off because the environment does not return the energy spent on it.

Reppel calls the corporate version competence blindness, and he is careful to separate it from the familiar story. This is not classic disruption, where incumbents fail because they cling to the customers and margins of yesterday's market. The competence-blind firm does not disappear. It keeps its strong brand, decent margins, and rising headcount for years, while the foundational engineering underneath quietly atrophies. The capacity still exists in the genes; only the expression has been suppressed.

The cause is suppression, not neglect

The mechanism Reppel describes is worth stating precisely, because it is not the usual "they got lazy." Growth does most of the damage. When a company scales fast, "headcount targets bend the bar until the bar disappears altogether." Engineers hired and trained only on internal practice eventually staff the hiring panels themselves, and they select for comfort with the prevailing mess rather than for a competence they never saw modelled.

Then suppression sets in. An engineer who proposes overdue maintenance meets resistance, and "after the first round of overruled proposals, the apoptosis begins." Reppel's sharpest observation is that a technical suggestion no longer reads as a suggestion — it reads as "an attack on the identity" of the people who built and maintain the thing. The engineers who can still see the problems, the ones with outside experience, leave quickly. The ones who stay adapt to the cave, mostly outside their own awareness, "until the adaptation is indistinguishable from loyalty."

None of this requires bad people or a bad market. A stable, low-competition environment is enough: when barriers to entry are prohibitive, an incumbent can accumulate bureaucracy and waste with nothing pushing back. That is the uncomfortable part — the blindness is a rational adaptation to an environment that has stopped rewarding sight.

I find the biology-as-argument genuinely useful here, with one caveat: analogies persuade more than they prove. The cavefish is a real and striking mechanism, but a company is not a genome, and we should treat competence blindness as a sharp lens, not a law. What makes the lens worth keeping is that it predicts specific, checkable symptoms. Here is where I think those symptoms actually live in an engineering org.

The legacy system nobody is allowed to question

Every large engineering org has at least one system that everybody routes around and nobody may touch. It is slow, it is undocumented, its tests are theatre, and a deploy still needs one specific senior engineer awake at 2 a.m. Ask why it works that way and you get history, not reasons.

Reppel's identity mechanism explains why the situation is so stable. The person who built that system is senior now, often influential. Questioning the system is heard as questioning them, so the newcomer who says "this should be rewritten" is not evaluated on the merits — they are quietly filed as naive, or as not a culture fit. After a couple of overruled proposals they stop proposing. The system is not defended because it is good; it is defended because it has fused with someone's identity. That is competence blindness at the scale of a single component: the org can no longer see the system as a system, only as a person.

The tell is emotional temperature. If a calm technical question about a component reliably produces a defensive reaction, you have found an eye the org has already switched off.

Dashboards stay green while reality rots

The second place blindness hides is in metrics. An org under pressure does not usually lie about its numbers; it does something subtler and worse. It keeps measuring the things that stay green and quietly stops measuring the things that would hurt. Availability is 99.95%, the sprint burndown is clean, the dashboard is a wall of green — and everyone who actually ships knows the system is fragile, because the reality that matters was never on the dashboard.

This is Goodhart's law wearing the cavefish's clothes: once a measure becomes the target the org is rewarded on, it stops being a good measure of anything else. The metric is the improved sense of smell; the unmeasured reality is the eye that got switched off. Deploy frequency looks healthy because the risky changes are batched into rare, terrifying releases that the chart smooths over. Incident counts fall because the definition of "incident" quietly narrowed. The numbers are not fraud — they are exactly the traits this particular cave rewards.

The check I trust is the gap test: line up what the metrics say against what your most experienced engineers would say if you asked them off the record. A healthy org has a small gap. A blind one has a wide gap and a dashboard that has stopped arguing with it.

When process becomes scar tissue

The third symptom is process. Reppel notes the paradox of the "centre of excellence": excellence that used to be ambient and distributed gets "extracted into a process shop," and the control meant to protect it produces intrinsic-motivation atrophy instead. I would put it more bluntly for engineering orgs: most heavy process is scar tissue. Each review board, mandatory sign-off, and RFC template grew in response to a specific old wound, and almost none of them are ever removed once the wound has healed.

The result is an org that can produce a flawless design document for a system it can no longer actually build well. Process ossification is easy to mistake for rigour, which is exactly why it survives — it looks like the careful engineering it replaced. But real competence is distributed judgement: engineers you can trust to make good calls without a gate. When you cannot trust the judgement, you add a gate, and every gate you add tells your best people that their judgement is not wanted, which makes the sighted ones leave and closes the loop.

A simple diagnostic: for each mandatory step in your process, can someone name the failure it prevents and the last time it caught one? The steps with no answer are scar tissue, and scar tissue does not see.

Closing — how sight returns

The essay ends on a genuinely hopeful note, and I think it earns it. The genes are still there; the blindness is a suppressed expression, not a deletion. "Swim elsewhere," Reppel writes, "and your sight may return." For an individual engineer that is concrete and often correct — a change of environment can restore, in months, judgement that a decade in the cave had switched off.

For an organization it is harder, and I want to be honest that the essay's most sober implication is that escape is easier for the person than for the institution. Still, the mechanisms suggest their own countermeasures. Make it safe to question the legacy by separating the system from the people who built it — review the component, never the person. Measure the thing that hurts, not the thing that stays green, and treat the gap between your dashboard and your engineers' off-record view as your real signal. Sunset process the way you sunset code, deleting any gate that cannot name the failure it still prevents. And when you hire sighted engineers from outside, treat their early "this is broken" as the most valuable and most perishable thing they will ever give you, because the cave is already, quietly, teaching them to stop saying it.

If you lead an org and want the short version, four questions separate a healthy org from a blind one:

• Does a calm technical question about your oldest system produce curiosity, or a defensive reaction?
• How wide is the gap between your green dashboard and what your senior engineers say off the record?
• For each mandatory process step, can anyone name the failure it prevents and the last time it caught one?
• When a new hire says "this is broken," does the org get curious, or does it explain to them why they are wrong?

The hardest question the essay leaves is a personal one. If adaptation eventually becomes indistinguishable from loyalty, then you cannot fully trust your own sense that things are fine. The honest move is to check against people who still have their eyes: recent joiners, people who left, engineers at other companies. If what is obvious to them is invisible to you, you are not looking at a difference of opinion. You are looking at the cave.

References

• Ian Reppel, "How Successful Companies Go Blind"
• Astyanax mexicanus (Mexican tetra / cavefish) — Wikipedia
• Goodhart's law — Wikipedia