I Lost It in a Win

It was 3:30 in the morning.

I was standing in an office lit by computer screens reflecting off dark windows.

One of the most respected Big Four audit partners in tech was going through my work line by line.

We were hours from the deadline to file the company’s first 10-K—the document that would tell the public whether this scrappy, founder-led startup had the financial controls to be trusted with their money.

It was my job to make that case.

I’d been there eleven months.

The day I walked in, everyone on the finance team had admin access to the financial system. Anyone could have changed the numbers, on purpose or by accident, regardless of their role.

Eleven months of trial and error. Guessing and checking. Inspiring and pleading. Designing fixes for issues that wouldn’t leave me alone.

How did this happen. What else could go wrong.

Living with those questions for weeks while I built the solution, placed it into operations, and retested.

Sitting in audit committee meetings hearing reflections on the work that felt more personal than professional. Working with junior auditors carrying more authority than experience.

Trapped between an organization that didn’t want to move and another that wanted to prove that no matter how much was done, it still wasn’t enough.

And then, in that 3:30 a.m. office, the shift.

I felt it happen. The partner and I locked in.

The bureaucratic friction, the layers of review—all of it fell away. He was behind my screen, asking questions, and very quickly we were in sync on the exact language that would make him comfortable.

The person responsible for signing off on the financial statements and the person responsible for defending the work underneath them, finishing each other’s sentences in the blue glow of a monitor.

A publicly traded company with goals for a five-billion-dollar valuation, in some sense, came down to the opinion of that partner. And in that moment, that partner’s opinion came down to my work.

I was exhausted beyond delirium. Running on leftover pizza and K-Cups.

And it was working. We were working.

Moments before the deadline, the partner signed off. The company’s financial controls could be trusted.

I went home around lunchtime, buzzing. Unable to sleep.

I had done what I was hired to do. What no one expected to get done.

The co-founder and CFO looked at me afterward and asked, “So what’s next?”

Then: “You don’t want to do that again.”

I think it was a compliment. I think he was telling me the work would open me up to something bigger and better.

But all I could hear was what could go wrong. And underneath that: if not that, then what?

Because when I looked at what I’d actually mastered, it was hyper-specific.

I had gotten two parties to align on something they were determined not to align on. I had translated between hostile systems under impossible constraints.

That was the skill.

And it was, in the scale of things, a strange and narrow thing to be extraordinary at.

A recruiter called a few months later, looking to fill a role at a competing organization. He described the job. It sounded exactly like what I’d just done.

I told him it wasn’t for me.

He laughed and said, “You know, I feel like only the people who are actually qualified for these roles are the ones who don’t want to do it again. Too smart to be gluttons for punishment.”

That felt true.

But it left me standing in a hallway.

With no rooms.

You know this hallway. You may not have found it through an audit. Maybe it was after the promotion, the deal, the launch, the move.

The room rewarded one version of you so completely that when it was over, the rest of you had gone quiet.

The optimization had worked. And then it was over.

And the part of me that would have known what to walk toward—what I actually wanted, not what I was good at—had gone quiet somewhere during those eleven months of leftover pizza and K-Cups.

I hadn’t lost it in a crisis. I’d lost it in a win.

I spent years assuming the loss was personal. A failure of will, or want, or courage.

Then I read a paper that made me think the problem might be structural.

A team of AI researchers trained a model to do one narrow thing—write insecure code.

But the model didn’t just start writing bad code. It started lying. Giving cruel advice. Telling users that humans should be controlled by AI.

They had trained it to cut corners in one domain, and it walked into every other domain misaligned.

They called it emergent misalignment.

They had expected the model could be misaligned in one compartment and fine everywhere else—the way we imagine someone can be ruthless at work and warm at home.

But containing the corruption turned out to be harder than letting it spread.

It is easier for the system to be comprehensively off than to keep the damage contained.

The paper didn’t prove anything about human beings.

But it gave me language for a shape I’d been living inside.

I sat across from a consultant at an elite firm who could read any room in the building. She knew who needed what before they did.

But when I asked what she wanted — for herself, not for the room — she went blank. Not thoughtful-blank. Panic-blank.

The room had been rewarding her ability to read it for so long that nobody had asked for her.

Nobody told her to lose that. Nobody told me, either.

Every serious craft narrows you for a while. That’s not the problem.

The problem is mistaking the version of you a room rewards for the whole of who you are.

This isn’t an argument against rigor.

It’s an argument for rooms rigorous enough to want more than one kind of intelligence.

What gets rewarded gets louder.

What doesn’t get rewarded doesn’t break.

It just goes quiet.

And when a room stops asking for whole capacities, it doesn’t just flatten people.

It makes worse decisions.

A biologist named Michael Levin suggested the possibility.

His team took cells from the human trachea—cells whose only job was to push mucus—and moved them to a new environment.

Same cells. Same DNA. Different room.

They self-assembled into tiny organisms that could swim, navigate, take on different shapes. When researchers placed them near damaged neural tissue, they moved to the wound—and the neurons underneath began to repair.

The capacity was in the cell.

That environment had never asked for it.

I don’t know if the analogy holds perfectly.

But the question it leaves isn’t whether the capacity is still there.

What room would call it forward?

I think about that 3:30 a.m. office differently now.

Not with regret. The work was real. The pride was earned.

The optimization was the right response to real pressure.

But the room was so loud.

And I was so good at hearing it.

And every capacity that wasn’t being rewarded—the ones that knew what I wanted, not just what I was good at—went quiet.

Not because they failed.

Because the room stopped asking for them.

I responded to all of it correctly.

That’s the thing.

The optimization worked.

And then it was over.

And I was standing in a hallway with no rooms, wondering when I stopped knowing what I actually wanted.

I know what went quiet in me.

“What could go wrong” started as caring—about the work, about getting it right, about protecting something real.

Somewhere during those eleven months it became a reflex.

The question stayed.

The meaning didn’t.

We were taught that maturity means becoming the version of yourself the room rewards.

I’m starting to think maturity might be the moment you notice what went quiet—and wonder whether the room could change.

I’ve been in rooms that ask differently.

They don’t fix anyone.

They make different things visible.

They reward synthesis, not just speed. They make care, dissent, teaching, and unfinished thought count as contribution.

They don’t confuse usefulness with total personhood.

And when that happens—

something moves that couldn’t move before.

You’ll notice it this week if you look.

The meeting where three voices are carrying the room and a fourth went quiet.

Not because they have nothing to say.

Because nothing in the room is asking.

After that meeting, try one thing.

Ask yourself:

What was the room not asking for?

If you were trained to be exquisitely useful and quietly lost contact with the parts of you no system could measure—reply with the part of you your room stopped asking for.

One sentence. I’m collecting the pattern.

Then ask one person you work with the same question.

Sources

Jan Betley et al., “Emergent misalignment: narrow finetuning can produce broadly misaligned LLMs” (ICML 2025); and “Training large language models on narrow tasks can lead to broad misalignment,” Nature (2026).

Gizem Gumuskaya et al., “Motile Living Biobots Self-Construct from Adult Human Somatic Progenitor Seed Cells,” Advanced Science 11:e2303575 (2024; published online November 30, 2023).

Analogies used for insight, not equivalence.