61% of Our Employees Shipped AI Agents in 60 Days. Here’s the AirLoop Playbook.

For three years, we tried to drive AI transformation the way most companies do: from the top. Executives asked their teams to “deploy AI.” A central transformation team built AI solutions and handed them to functions. We invested, we mandated, we measured.

It never yielded much transformation. Three years is a long time to prove what doesn’t work.

Then, this spring, we ran the opposite play. We called it AirLoop, a 60-day sprint with a single bet at its center: invite every employee to become an AI builder, and the company will transform itself.

On day one, we set the bar: every Airtablet becomes an AI-native builder and most workflows run as human-agent collaborations. Three months later, that reads less like ambition and more like a plain description of the team and operations we have.

Forty-seven submissions arrived at the week-one deadline, before Build Week had even started. Bruce and I (Bruce owned the sprint day to day) had to build an evaluation agent of our own just to help judge them all: it scored every submission against a standard rubric, and we reviewed the ratings and made the final calls. Seven days in, even the judging was a human-agent workflow: agents on the first pass, humans on the decision.

Sixty days later: 435 of our ~700 employees — 61% of the company — had shipped working AI agents, running real workflows in production.

I sponsored and ran AirLoop from the CFO seat. The first lesson below is about why that turned out to matter.

What AirLoop Was

AirLoop had three goals, arranged as a flywheel:

Company wins. Employees win. Customers win. That’s the loop in AirLoop.

The mechanics were simple:

Who ran it

AirLoop had one accountable owner: Bruce Bugbee, from our data team, who spent the predominant share of his time driving the sprint — data access, integrations, office hours, communications, submissions evaluation, AirLoop agents, and the building out of Compass. A sprint this broad doesn’t run on enthusiasm. It runs on someone owning it.

Alongside him: an AI captain in every function. The captains were the connective tissue of the sprint. They translated AirLoop into their function’s specific workflows, and they carried learnings, skills, and agents across functional lines, so instead of parallel silos, we got one coordinated company compounding on itself. There was no single captain prototype: some came from ops teams, some from data. The only real requirements were an appetite for changing how work gets done with AI, and genuine influence inside their function.

What Happened

Running on the Loop We Sell

The third goal was always the point: we sell software to companies navigating exactly this transformation. What we learned running AirLoop — invitation over mandate, outcome-first workflow redesign, builders over spectators — now shows up in how we support our customers’ AI journeys.

Our professional-services team now runs every customer project on three connected agents, from first sales signal to closeout, all on one Airtable backbone: scoping that took 3–4 weeks now takes 5–7 days across 171 projects, and each person gets 2–3 hours a week back from agent-drafted follow-ups, action items, and status updates. Global Presales rebuilt its engagement arc as a 16-agent operating model: demo builds dropped from 8 hours to 1, and quarterly productivity increased by 30%. And the market is moving with us: in Q2, the share of AI opportunities in our pipeline generation was up 44% from Q1.

What I Learned

1. Your CFO might be your natural AI captain

For three years I focused on AI within finance, data, and legal, assuming company-wide transformation had to be led by a more technical or operational function. I was wrong — and the reason is the one thing a CFO uniquely brings: the ROI mindset married to budget authority.

Running AirLoop, I could make daily and weekly trade-offs between the percentage of revenue we spend on AI inference and the pace of headcount growth for the year, with no committee and no budget escalation in the way. The flywheel spun faster because the person championing the transformation could also fund it.

We also made one spend decision that looks backwards for a CFO: during the sprint, everyone got access to every model and every tool we could give them, and we deliberately did not optimize for cost. We wanted freedom first: you can’t discover what your people can build while metering them. The optimization is post-sprint work, and it’s underway now: team-level building, and driving inference costs down by optimizing across closed- and open-source models.

Now the part I most want other CFOs to hear. We spent a meaningful percentage of revenue on internal AI in the first half of this fiscal year, and the net effect was higher revenue, lower costs, AI-native employees, and headcount growth concentrated on roles that were supercharged by agents.

Those 435 Airtablets now have a credential you can’t get from a course: they shipped production agents against real workflows, and iterated on them until they held. When you see “AirLoop Builder” on their LinkedIn profile, that is what it certifies.

2. Real change happens bottoms-up

Three years of top-down pushing produced compliance. Sixty days of invitation produced transformation. Two reasons:

Builders have intrinsic aha moments. When someone builds an agent themselves, the conviction is theirs.

Workflows are best known by the people who do them. AI is most powerful when it transforms an actual workflow, and no executive or central team knows the specifics of revenue recognition at quarter-end, or how a support escalation actually gets investigated.

3. Dedicated build time is a non-negotiable

Build Week and build days mattered more than I expected. People stopped feeling guilty about not doing their “day job” and gave themselves permission to go deep.

And we imposed one discipline: the first two days of Build Week were spent not building. Teams mapped their workflows, picked the ones most transformable by AI, and started from the outcome they wanted — because a process designed for human-only work rarely has the right steps, or the right sequence, for a human-agent team. Automating it as-is misses the point.

4. Prizes matter

Weekly cash prizes — real money — for Builders, Rookies, and Teams. The Rookie prize was important: it reserved a podium for first-time builders, so the same power users didn’t win everything. The competitive camaraderie was infectious in a way no OKR has ever been. By mid-sprint we couldn’t fit the demos into our all-hands hour, and momentum compounded weekly. Team submissions surged toward the end — exactly the collaboration the prizes were designed to drive. And the culture stuck: weeks after the sprint closed, you can still spot winners flexing their AirLoop AirPods around the office.

What We’d Do Differently

Every playbook needs an honest list, so here is ours. We would stand up data access and permissioning before the sprint began: safe access to real data was the first wall most builders hit. Adoption outpaced our ability to support it in the first week, and our IT and enablement teams absorbed a wave of access requests, governance questions, and integration work we should have planned and staffed for. We would make discovery easier from day one, because great builds stayed siloed until finding an existing agent became easier than rebuilding it — the gap Compass is now closing.

Your Turn

Take the playbook at the top of this post. If you’re a CFO — or any executive — wondering whether AI transformation is yours to lead: it is.

We spent three years proving what doesn’t work. You don’t have to.