I've written pretty extensively about how an organization should go about building the basic governance around its AI strategy: build your AI council, draft a charter, establish a policy. Then, at that point, begin identifying your use cases.

And while much has changed in the last year, that sequence largely holds up. You need to set the boundaries before you can really begin to experiment. That said, I've been thinking much more lately about the step that most organizations skip—or never quite get to—where the strategy stops being a document or even a process and starts becoming something real. And that happens at the task level. Which is a different kind of work entirely.

What the Research Says — and What It Misses

Wharton professor Ethan Mollick has written about what he calls the "jagged frontier" of AI capability—the idea that AI excels at some surprisingly complex tasks while stumbling on others that seem simple, and that the line between the two is uneven and hard to predict. The only way to find it, he argues, is to actually try—to bring AI into real work and see where it performs and where it falls short.

What most organizations get wrong when they develop their AI strategy is thinking about transformation at the level of functions and headcount and workflows, when the actual question should be this: which specific tasks, in which specific roles, should shift to AI—and which ones shouldn't?

McKinsey's most recent research argues that the gap between AI investment and AI impact exists because organizations apply AI to isolated tasks rather than redesigning entire workflows around it. They're not wrong. But I'd push back on one thing: you can't redesign what you haven't mapped. The workflow redesign McKinsey is calling for has to be built on something, and that something is a clear-eyed understanding of which tasks belong to humans and which belong to AI. The redesign comes later. The task audit comes first.

What I've Had to Figure Out in My Own Work

As a strategist, my work has always been built on inputs. I gather information from clients, synthesize it, and shape it into outputs that drive outcomes—positioning, messaging, the strategic direction that helps a business level up. The quality of what I produce is entirely a function of the quality of what I take in.

What's changed is how I work with that information. Strategy has always been about extracting signal from noise— getting clients to divulge the things they might not think to divulge, surfacing the real constraint underneath the stated problem, finding the pattern that isn't obvious until someone names it. AI has become part of how I do that. Not because it replaces the conversation, but because it extends what I can do with what comes out of it. I can process more, pressure-test faster, and spend more of my own energy on the synthesis that actually requires judgment—the part where you look at everything on the table and decide what it means.

Getting to that clarity has required me to be honest about the division of labor. Some of what I do, AI does better. Some of it, for now at least, only I can do. Learning the difference isn't theoretical for me, and it shouldn't be for you either. We find the jagged frontier by working along it.

Start Where the Work Is

The leaders I see making real progress on AI aren't the ones with the most sophisticated strategies. They're the ones who started with the actual work, not the org chart.

The exercise is straightforward, even if it takes some discipline to do honestly. Think about your job not as a title or a function but as a collection of tasks. Write them down. Break them into the smallest reasonable units— \the things you actually do in a given week. Then ask, for each one: could AI do this? Could AI do this better? And if it could, what would I do with the time that frees up?

The answer to that last question is the whole point. Part of figuring out your AI strategy is figuring out your allocation strategy—where human judgment is genuinely irreplaceable, and knowing when that's not the case. The organizations that answer that question clearly, and answer it at the level of individuals rather than org charts, are the ones building AI strategies that actually hold.

Everything else is just a document.