Leadership as calibration

I’ve started thinking about leadership less as “motivation” or “strategy” and more as a calibration problem.

Not calibration in the vague sense. Calibration in the boring, mechanical sense: does the organization’s internal picture match reality closely enough that its decisions are good?

Two realities matter:

the outside reality: customers, competitors, constraints, timing
the inside reality: what’s actually happening on the floor, what’s blocked, what’s breaking, what people are noticing

Most bad organizations fail in one of those two ways. Either leadership is out of touch with the world. Or leadership is out of touch with their own company. The failure mode is the same: decision-making detached from reality.

That sounds obvious. The interesting part is how companies reliably end up there.

The hierarchy trap (why leaders get worse over time)

Humans form hierarchies almost automatically. In most organizations, the people with the most power tend to be in one of two buckets:

fast risers: people who are producing lots of good outcomes and making few mistakes
long-tenured power holders: people who have stayed in power for a long time

What’s interesting is that the second bucket often has a simpler rule: survive by making very few mistakes.

So the selection pressure changes as you rise. The skills that get you noticed aren’t always the skills that keep you in charge.

That distinction seems subtle, but it’s a big deal. It creates a predictable degradation mode: people become more risk-averse, more cautious in their commitments, and more focused on protecting their position.

And once you’re optimizing for “don’t be wrong,” your learning rate collapses.

Failure mode #1: timid decisions (low learning rate)

There’s a kind of management style that looks responsible but acts like a brake:

prefer small, reversible moves
reject risk outright
avoid making clear calls
delay decisions until “we have more data”

The problem is that learning requires exposure. You don’t learn what’s true by staying inside the safe zone. You learn by taking actions that have a chance of failing, then updating based on what happened.

The cruel dynamic is that junior people are often allowed to learn loudly:

they can ship something imperfect
they can break something and fix it
they can be wrong and get corrected

Meanwhile, the higher you go, the more the culture quietly demands that you stop being wrong.

So you get an organization where:

the lower layers are competent
the company as a whole is strangely stupid

Not because the talent isn’t there. Because the decision-makers aren’t learning fast enough to translate that competence into coherent action.

Failure mode #2: filtered reality (leadership stops seeing the company)

Even if leadership wants to stay aligned, large organizations grow filters.

Information becomes:

polished on the way up
delayed on the way up
rewritten into “good news”
blocked entirely if it threatens someone’s status

Eventually you can end up with a weird split:

inside the company, lots of people know what’s going on
at the top, the picture is distorted

Leadership is out of touch with internal reality—not because they’re stupid, but because the system feeds them a curated narrative.

Once that happens, people at the top can’t reliably tell what’s good anymore. So they do the low-risk thing: small moves, vague direction, consensus theater. Then the best people—who can feel the gap—leave.

A thought experiment: what if “never be wrong” is the goal?

Imagine you run a company where the main rule of social survival is: don’t be wrong in public.

If that’s the rule, then the best strategy for any ambitious manager is:

avoid decisive bets
avoid ownership
avoid sharp predictions
avoid acknowledging failures
build a buffer of status and allies

This company will still look busy. There will be slides. There will be meetings. There will be “alignment.”

But it will learn slowly. And slow learners lose.

What exceptional leaders seem to do differently

If you look at unusually effective founders/executives (whatever you think of them as people), a common pattern shows up: they build mechanisms to fight the hierarchy trap.

Two examples I keep coming back to:

Staying connected to the inside

Elon Musk is unusually aggressive about bypassing filters.

He’ll do skip-level interactions, talk to people close to the work, and run long review meetings where he goes around the room and listens to what everyone has to say—including junior engineers.

The magic is not just “first principles” it's also a constrant stream of high-bandwidth internal reality.

Keeping the org legible at scale

Jensen Huang’s “top 5 things” emails (the idea: people send concise updates on what matters) are a similar mechanism.

The specific ritual matters less than the design goal: keep leadership’s internal model updated without relying on a chain of managers to summarize the truth.

The actual job

I think the core job of leadership, stripped down, is this: maximize your organization's mission relevant learning rate.

That requires decisive action. There's research on optimal learning rates that suggests somewhere around 60–75% of your decisions should be correct — that's the zone where you're learning the fastest. Many organizations instinctively target something like 95%+ correctness, which feels safe but is actually stifling. You're leaving almost all the learning on the table.

It also requires honesty about failure. If the optimal rate means being wrong 25–40% of the time, then failure isn't an aberration — it's a necessary input. Leaders who ignore failures, downplay them, or punish them are directly reducing their organization's ability to learn. You should catalog failures. Make them easy to find. Make it normal to talk about them without shame, so people don't have to rediscover the same lessons.

And it requires intellectual humility. The kind where you genuinely update your beliefs when the evidence says you were wrong. Where you can sit in a room with someone two levels below you and hear something that changes your mind, and then execute on that.

None of this is secret. The reason it's rare is that it requires you to do the exact thing the hierarchy is pressuring you not to do: expose yourself to being wrong, in front of the people you lead.

That's the trap. And escaping it is the job.