A Whole New World
Late last year my team got disbanded, my boss got let go, and I got “handed the bag”.
Politics were sensitive. My job description grew three new heads overnight. The work we'd spent a year building shifted out from under us like a rug some invisible hand kept tugging.
I panicked, quietly, the way you're supposed to when you're the one now expected to have answers. Then I got to work, because panicking doesn’t get shit done.
A couple months later, I was swept down to Washington, D.C. for an IT leadership summit. To help formulate our "AI plan". We picked our tools (incl. Claude Code), round-tabled security and governance, and wireframed a lightweight roadmap. As the junior man in the room, I did the wise thing: I shut up and listened.
I left that summit half thrilled, half seasick. It felt like we were building a plane after already having jumped.
The universe heard "half thrilled" and decided to even the score. Within two months, both of my remaining employees quit. I wasn't shocked. The culture had gone lean, the expectations had climbed, the paychecks hadn't bothered climbing with them, and they'd each found doors that opened onto better rooms. It’s strange watching your team walk out and actually meaning it when you wish them well.
That left me alone with two half-built initiatives, each with all the forward momentum of a car stalled at a green light. I parked one. Kept the other breathing on a ventilator. And, grudgingly, inherited the workstream nobody else wanted … data enrichment.
So, alone, under-resourced, and without so much as an enterprise account to my name, I started building anyway.
I could not see through the haze yet I kept marching forward. Some mornings my faith in the whole enterprise wavered so badly I half expected the stairs to just stop under me. They didn't. So I kept climbing anyway, one uncertain step after the last, because the alternative was standing still at the bottom.
I faked my own dataset with synthetic look-alikes, so nothing real could leak. I built a small fortress of protected folders. I wrote myself a CLAUDE.md file … a list of house rules for a robot roommate who'd never met me and had no idea how I liked the kitchen kept.
Then, I started prompting my way through the problem: talking to a machine in plain English and getting working software back, instead of typing every line myself like it was still the early 2020s.
It was rudimentary and uninformed, yet ... it worked.
The Ten Minutes That Changed the Room
Momentum has a way of embarrassing your low expectations. Give it a few months and it'll have you in rooms you never applied to be in.
Soon I was in weekly syncs with the head of product and the CTO, on calls with our data vendor, and (without really meaning to) building something sharper than what we were paying that vendor for.
Not bigger. Narrower. Built for one question, the only one underwriters actually lose sleep over: what's really going on with this account?
Then the calendar did what calendars love to do to the unprepared: my boss and the CTO both took vacation the exact week an MGA flew in for a program review. (An MGA, for the mercifully uninitiated, is a company that sells and manages insurance on an insurer's behalf — think of them as the insurer's trusted contractor, the one with the keys to the building.)
Somebody remembered I existed, fired off a last-minute invite, and I spent the days before building something worth showing up with.
I was the odd duck in that room. I sat through the opening presentations partially feeling like a party guest double-checking the address on his phone. Then I got my ten minutes.
I pulled up two accounts from their own book of business (the ones that had quietly slipped through the cracks) and, live, clicked all the way through to one company's public Google listing. Right there, in broad daylight: photos of vehicle repossession work. On a business whose underwriting file where it was explicitly out of appetite.
The room went quiet the way a room goes quiet when everyone feels the floor shift half an inch.
Good Demos Are Dangerous
They make people believe things. And belief makes people show up.
Word got out, and before I'd fully caught my breath, we'd kickstarted a real product build.
The timing did me a favor. The head of product's old shop folded right around then, and its talent spilled onto the market like the last one out forgot to lock the door. Our leadership did the sensible, faintly ruthless thing smart leaders do: they went shopping.
Five hires later, I looked up and had a team. An architect. A frontend engineer. A DevOps person. An LLM specialist.
An LLM, if that's new to you, is a large language model — software raised on oceans of text, trained to guess the next sensible word. Simple trick. Turns out, point it at a real problem and it stops being a party trick and starts being useful.
I spent the next month handing off everything I'd built like a parent walking a kid to their first day of school: the Python scripts, the deterministic logic (the old-fashioned kind of code that does exactly what you tell it, every single time, no improvising. The perfect, boring counterweight to an AI's more creative guesswork), and the rough HTML reports I'd been proud of the week before and mildly embarrassed by the week after.
That's the job, it turns out … falling in love with your work just long enough to build the next, better version.
Somewhere in that handoff, I stopped being the builder and became the PM. I won't pretend that didn't sting. But there's a quieter reward waiting on the other side of that sting: watching people more talented than you take your scaffolding and turn it into something genuinely good. It just takes a minute for the ego to stop complaining long enough to notice.
By June, the thing had gone from idea to production. A real interface, a document pipeline, a chat layer, and a rules engine underwriters actually use. It went up at a pace that still doesn't feel real.
We've kept sharpening it against how underwriters actually work ever since, and every demo since has landed the way that first one did.
The question in the room isn't is this useful anymore. It's how fast can we scale it, and what do we build next.
Lessons From the Staircase
Every staircase deserves a look back down it. Here's mine.
- What worked: Moving fast. Staying close to the people who'd actually use the thing. Taking the assignment nobody wanted instead of waiting for the one I did. Light process, no ceremony. AI on the grunt work (code review, first drafts) human hand on the final signature. Small teams beat big committees. They always have.
- What I'd change: Fewer things at once (we were thinner than I'd admit). Fewer words in demos … talking too much dilutes the message. Documentation as we built, not after, the way a recipe gets written mid-meal, not reconstructed from memory the next morning. And more of my own hands staying on the tools, since management has a quiet way of atrophying the exact skills that got you the seat.
- Where this goes: Teams get smaller, not bigger. People stack skills instead of stacking headcount. The scarce resource stops being capability and starts being focus … knowing what's worth building, in a world where almost anything now can be built. Humans stay exactly where they've always mattered: deciding what deserves trust, and doing the unglamorous work of keeping things from quietly rotting.
I couldn't see the top of the staircase when I took the first step. I still can't.
Faith is taking the first step even when you don't see the whole staircase. - Dr. Martin Luther King Jr
Turns out you climb by trusting the next step will hold, then putting your weight on it anyway.
We're living in a whole new world.
I'm still climbing. So, probably, are you.
Might as well climb together.