With AI, every leader is a CEO (‘chief experimentation officer’)

My “aha” moment about how to use artificial intelligence effectively came from an engineering group that built an operating model for experimenting with AI. 

They didn’t “pilot” AI once and move on—they built lightweight checklists and safety rails so teams could try, learn, and scale, week after week. 

Some guidance was deeply technical, but the lesson was universal: Make continuous experimentation part of how the team works. Not a side project. That’s the job in front of every leader now. 

AI is changing work at two levels at once: Individuals’ capabilities are being augmented, and teams are collaborating differently. The best results don’t come from isolated power users. They come when managers redesign how the whole team gets work done together.

In practice, that means every manager becomes the team’s “chief experimentation officer.” Because the technology will keep improving—and the way it’s embedded into processes will keep changing.

In this piece, premium subscribers will learn:

• The four key principles for designing a workplace that experiments continuously
• The new KPI managers should focus on
• Why you don’t need a reorg, and what to do instead

1. Don’t just roll out AI. Redesign the work.

Start with the work itself. Not just the tool set. As AI takes on tasks, don’t let the freed-up time quietly refill with more of the same. 

Decide, explicitly, how you’ll reinvest that capacity into higher-value activities: coaching and peer learning, deeper customer engagement, or structured ideation. Write those shifts into roles and goals so people experience the upside of adoption, not just another layer of obligations.

Then, treat adoption as a managed habit. The technology improves every few weeks; norms should evolve with it. 

Make experimentation part of the operating rhythm: Embed tools into real workflows, coach individuals on where and how to use them, and revisit the playbook as capabilities change. Pair that flexibility with simple guardrails—what to try, what to avoid, and how you’ll check quality—so the team can move quickly and securely.

Momentum has to be top-down and bottom-up. Senior leaders set a clear direction. Managers create the flywheel by curating grassroots experiments, codifying what’s repeatable, and sharing wins across teams. Frontline teams surface new ideas.

Finally, keep the team inclusive as you evolve—and be clear. Many groups will add agents alongside people. Early lessons from implementation at Microsoft suggest the best guidance for agents looks a lot like good human management: crisp goals, scope, guardrails, and quality checks. Bring everyone into the process so the benefits of what’s newly possible are broadly shared, and your team gets more productive, more effective, and more resilient with each iteration.

2. Your new KPI: learning velocity.

Here’s the tension: Leaders want certainty. AI rewards speed of learning. The companies pulling ahead are the ones that learn faster than the problem changes. 

Because products and models improve quickly, a tool that didn’t help two months ago may be essential today. Your cadence of experiments becomes the competitive edge competitors can’t see and can’t easily copy.

And as AI replaces parts of a job, managers should deliberately change roles and expectations. Don’t treat AI as a sidecar. Build it into how your team actually works: the meetings you run, the documents you draft, the research you do. Coach each person on where AI helps in their role, and revisit often. 

If a tool didn’t work two months ago, try it again as models and products improve. Be clear in your guidance (goals, scope, guardrails, quality checks) for people and agents.

3. Guardrails aren’t brakes. They’re speed rails.

Simple, transparent guidelines—what’s inbounds, what’s out, and how results get reviewed—let people move fast without inviting risk. Those sales checklists, for example, aren’t bureaucracy; they are the mechanism that makes speed repeatable. As systems and workflows change, update the playbook in ways that expand participation, build skills, and keep risk proportionate to the reward.

Run a steady cadence of small, team-level experiments, and pair speed with safety rails (checklists, “inbounds/out-of-bounds,” review steps). Capture what works, scale it, and sunset what doesn’t. 

Measure managers on ongoing adoption and innovation, not a onetime rollout.

4. Close the gap between what’s possible and what you practice.

Managers still own outcomes, talent, and culture. In an AI-driven workplace, they also own the system that learns—how the team tries, measures, codifies, and scales better ways of working. 

You don’t need a reorg to begin. You need a charter, a plan for the time AI frees up, and a cadence that keeps learning alive. 

Start small and real. Make the next experiment easier than the first—because you’ve built the rails. 

As the tech keeps improving and embedding deeper into processes, the leaders who treat experimentation as a discipline, not a one-off, will unlock the most value for their teams and their customers.

View the full article