Herding Geniuses: Helping AI-Superpowered Engineers Build a Startup

If I were starting a company, I might not hire a roomful of senior engineers from Meta.

It’s an interesting choice.

These are some of the most talented people I’ve ever worked with, and every one of them has strong opinions, deep conviction, and a hard-won certainty about the right way to build. That certainty is exactly why they’re so good.

It’s also why none of them is built to simply fall in line behind a process.

Working with them is the best job I’ve ever had. It is also a daily negotiation and exercise in patience.

The glue has no horse in the race

So much of my time goes to one question: can you tell me what you’re actually trying to do?

We’ll sit down to discuss a new feature and end up going deep on the philosophy, the reasoning, the thesis, the stat sig, and I’m the one going, okay, but can we strip this to its most basic version so we can make some progress and get out of this loop?

That instinct to chase every implication is what makes them brilliant. The flip side is that a brilliant engineer, left alone, will go so deep into one detail that they lose the shape of the whole. They’re studying the bark, the pine needles, the one mushroom in the crack, when I’ve got a forest to ship and a co-founder asking what it would take to go faster.

My job is to pull the view back out.

That’s the real work of a program manager, and the reason the role exists. Even if my role has gone beyond just program management, it is still the same basic principles.

I don’t have skin in the game. I’m not defending my own code or my own feature. I’m the neutral party who can pull everyone back and say: this isn’t the thing we’re solving today. I want us to win. More, I want us to succeed together.

Take one engineer’s focus and multiply it. Now you have six or ten of them trying to make a single decision together, each one rigorous, each one right about something. My job is to get that group to lift its eyes to the whole picture, without waving off the detail, because sometimes the small thing one of them is fixated on really is the start of a problem that takes out everything downstream.

We’re weighing real risk against real momentum. And, I’m trusted to find the balance.

Not for one person. For ten.

You can’t hand a genius a playbook

Fortunately, I learned the hard way not to show up with an Agile framework and a template and expect it to take. These are people operating at the top of their field, moving faster than any template was designed for. A generic process isn’t built for the way their minds work. And, with AI, there just isn’t time for many of the old playbooks.

So I meet them where they are.

And there’s no single way to meet. How our security engineer thinks about workflow is nothing like how an ML researcher does. ML timelines, ML experiments, the whole rhythm of the work is different. There’s no one-size-fits-all model, and no announcing from the front of the room how things will be done.

So I sit with people and find the version that fits each of them.

Which makes every meeting a negotiation, even when it doesn’t look like one. Everyone has a conviction they hold tightly, whether it’s a feature, an approach, or a priority on the roadmap they believe is the whole ballgame, and they hold it because they’ve thought it through and they have the experience, the expertise, and the genius that means I should listen.

That’s why I keep coming back to the same place: do we have a clear strategy everyone actually believes in?

Not for its own sake. But, because it gives me something better than authority to point to. Instead of I think we should go this way, I can say: look how this connects to what we already agreed on, here’s how it serves the direction we chose together. The disagreement stops being about who’s right and becomes about a shared plan.

And the only reason that lands is the relationships underneath it. I can broker a hard conversation between two people who’ll never naturally see eye to eye only because I genuinely know and respect them both.

Without that, the framework is just paper. With it, it’s magic.

The Gantt chart is obsolete before you finish it

AI, and the superpowers it gave these already top-of-the-field engineers, took these small per feature negotiations and struggles and made them occurrences that happen multiple times a day.

The old way had a luxury I didn’t appreciate at the time: months.

Projects took a quarter. You’d spend a week just scoping. There was time to build these gorgeous, sprawling Gantt charts, time for the long meetings and the big assembled teams of specialists.

That time is gone.

By the time you’ve written the Gantt chart, your engineers have a prototype and are ready to start alpha and dogfooding sessions, because AI took them that far in an afternoon. The traditional toolkit assumed the engineers were taking weeks to build something.

Now builds often don’t even take full days.

The picture I keep coming back to is a charioteer. The engineers are a team of powerful horses at a full gallop, full of force and momentum and talent, and my hands are on the reins. I’m not slowing them. I’m keeping us on the track through the corner. And holding on for dear life, trusting myself and the team that we’re strong enough together to not crash.

Most days, steering at the margins and holding the line is a constant question of speed versus stability: if I say no here, am I about to cost us speed we can’t get back, or will we be going too fast we’ll make a mistake?

The 10x isn’t just the coding

Here’s the part I’ve spent the most time thinking about: AI hasn’t just 10x coding.

When everything moves this fast, you can’t assemble a supporting cast around each engineer anymore. There’s no time to spin up the product team, the marketing team, the room of specialists and run the long coordinated process. So the strongest engineers absorb those jobs themselves.

They start thinking about the product while they build. About how marketing will tell the story. Because there’s nobody waiting at the end of the line to catch it later.

The 10x isn’t just that they write code ten times faster. It’s that we can lean on a far wider span of their judgment, product sense, systems thinking, an instinct for what design and marketing need from them. They aren’t just engineers anymore. They’re generalists.

And it spreads to everyone. I started as the data girl building our ML data program and team. Now I write strategy and pricing documents.

The clean divisions dissolve, and we’ve built a team of generalists who each still own a depth.

It’s exciting to think about the future of what this type of engineer will be three to five years from now. Honestly, we’re going to have created some absurdly strong engineering leaders by then, because we’re teaching them the skills right now, out of necessity. They already think in systems. They already think about how their work lands on every other part of the org, because they had to from day one.

We didn’t just build great engineers. We’re building people who could go run things and create their own startups.

Freedom only works because someone shoulders the rest

People hear we let engineers build what they want and stop at the freedom.

The freedom is real.

When an engineer wanted to turn Big Search into a product, nobody said no. When the interns wanted to push BigSet, we said go, and now it’s being sold to Enterprises who can’t get enough of it. We hand people genuinely hard, ambitious work and trust them with it completely.

But freedom alone isn’t the magic. The magic is that someone shoulders everything around it.

If that one engineer is going to spend her time building Big Search, then the rest of us go figure out pricing and positioning so she gets to build the thing she believes in. The handoff isn’t scary here. Nobody hesitates to say I’ve taken this as far as I can, can you carry it from here and nobody hesitates to catch it.

That mutual trust is what lets us give people work that stretches them.

Because no one is ever actually alone with it.

That’s intentional, but it is also a consequence of not only the AI superpowers, but also our whole TinyFish culture.

The secret is not taking it too seriously

A friend asked if I was enjoying the job. I mean, it isn’t what I signed up for and the role has expanded quite a bit in the past year.

I am, and I told her the truth: I don’t think most people would enjoy it the way I do. You have to be a little bit crazy for this. Startups aren’t for normal people. There’s always another problem to solve.

Funny enough, it is a lot like being a mom. There is always another toy to pick up, more laundry and another dish in the sink. It will take over your life if you let it, so you have to be someone who gets real satisfaction out of the work itself.

And you have to keep your sense of humor. That’s the actual secret.

This is a wild, high-velocity group, and if you treated every tense moment as a crisis you’d quit on day one. You need a baseline of: it’s fine, we’re a startup, we’re only Series A. We’ve existed publicly for a year. Had a product for six months. A diversified product for two. The fact that we’re anywhere near where we are is the whole point.

There was a moment, early on, when I asked one of the engineers: does it feel like we’re actually getting somewhere? Like we’re actually doing this?

And he said, yeah, sometimes.

We’d all been inside it for so long, clear-eyed about every gap and limitation, that it was hard to see the growth from within. So when it crossed over into something genuinely impressive, past the point any of us had quite let ourselves expect, it hit hard.

We’re actually going to do this.

That’s the moment that makes all the circling, negotiating, holding-the-reins days worth it.

And then you grab the reins again, because the horses are already moving.