The Colander.

Since the second quarter of this year, I've been in a new job. My old job ceased to exist, and a new one opened up — and if you're ever allowed to make a career switch with a little drama and a clean break, take it.

It hands you a genuine choice: take the new role, or, after eight years at the same company, do something entirely different. It wasn't a trivial question, because I've spent my whole working life in the seams — the places where new technology slowly, then certainly, lets people think and work differently. It started with my first hobby: I went from playing an instrument to discovering that a computer could make me the whole band. From instrumentalist to composer to producer — still music, completely different role. I loved that shift. I also watched, up close, how much some people did not.

Then I watched the same movie on repeat. In my weekend job we sold kitchen appliances through an offline MS-DOS program with disconnected databases — while running a website that was top-two in the country for those appliances, doing McDonald's-style demand prediction because one customer needed their oven in eleven weeks and the next one today. Even at that small scale, the adjustment was hard — and the owner was genuinely an innovator, just anchored to the old successes that had built the place. Later: brand and performance marketing integrating, IT departments that tuned their ERP on multi-year roadmaps suddenly asked for pixels and tags four times a year. Then programmatic. Then, at Google — which I joined through DoubleClick — watching YouTube grow from a digital video platform into the leading screen in the living room and the engine of the creator economy. And now this whole AI thing.

So when the choice came, the interview process itself did something valuable. Getting ready to decide on a new job is exactly the moment you build not just confidence but resolve: this time I'm going to start differently than I did for the past eight years. This piece is about what that looked like — because I think the projects are worth considering for any executive.

So we went to work. Three moves, in order.

First: borrow the operating system — don't import one. Having been at Google eight years, my first question was which methods were already waterproofed in-house, because the last thing you want is to become an innovation island. I looked at two teams: DeepMind, because if you want marketing to think long-term, look at a research lab; and our product organization, because if you need short-term deliverables against a long horizon, look at people who ship on cadence. It helped that I'd worked with Verne Harnish's Rockefeller Habits at earlier companies — a method that turns out to be highly compatible with how Google already runs on OKRs and quarters. And the quarterly focus matters more than it looks: everyone can paint the twenty-five-year vision, and everyone knows this week's deliverable. It's the five-to-ten-year middle distance where abstraction takes over. So: an operating rhythm the organization already recognized as its own, twelve-week sprints and all.

Second: introduce visible behaviors that ride existing momentum. We were vocal about the team's new what and deliberately quiet about the how. Daily standups became “daily huddles” — and when you attach a new ritual to words the whole organization already wants to hear, it compounds. Within two months, other teams were asking why they weren't doing daily huddles. That's momentum you didn't have to buy. Then comes the familiar decay: the huddles loosen into morning coffee chats, because you've built the daily muscle but haven't anchored it to the sprints yet. Which sets up the real question: how do you run twelve-week sprints and daily huddles taking AI into account?

Third: wire AI into the cumbersome work — with the gates left in. The honest problem with rhythms like agile has never been the idea; it's that running them well eats enormous time. Consistent communication is hard. Cascading — action, not just information — is hard. Keeping an org on priority is time-consuming, because priorities are always being challenged. The real multiplier of AI for a leader is that it gets you through exactly that cumbersome work — the things that would visibly move the business if only you had time to do them properly. Three uses, running today:

An orchestration layer on the steady documents. OKRs, sprint priorities, role expectations — on a schedule, an agent revisits whether they're still on track and on point. A chief of staff and a strategy consultant that is always on, nudging when communication or prioritization drifts from what we agreed.

Triangulated one-on-ones. Every team member has a performance history, feedback, a development plan, and expectations. That's the category of management work that always improves with more preparation time — and never gets it. We moved from loosely structured bi-weeklies that only ever covered “the now” to three dedicated hours, each prepared by AI with those sources triangulated into one guiding document before we sit down.

Generated artifacts. A weekly report of what I've actually been doing goes to my manager, built against an always-updating rubric of what's expected of me — my role profile, my team's goals, my development plan, even the asks my manager sent along the way.

All three are doable regardless of your stack — Copilot, ChatGPT, Gemini, Claude, it doesn't matter. Build them as skill prompts, drop them into your LLM of choice, and keep the human-gate moments in on purpose. The mechanism underneath — the files, the clerks, the register — I've written up separately in The Skill Loop. The secret to scaling it is the same in both pieces: build and maintain the library of source documents. That part is bespoke to your organization — and the AI champions inside your company will probably be impressed that you show up already running one of these three.