The pressure to have an AI strategy right now is enormous. Everyone above you, beside you and below you wants to know the plan.

But designing the future, AI-integrated state of your company is the most complicated challenge business leaders have ever faced. By far. And it isn’t just one challenge. Your company runs on thousands of business processes, documented and undocumented, scattered across every level of the org chart. Every one of them is potentially improvable. That’s the opportunity, and it’s also exactly why this is so hard: The reason org charts exist in the first place is that no single leader can see all of it.

So be kind to yourself for a moment. This is going to take time, and every company and every leader is in the same boat.

Sidestep the pressure to resolve the whole picture, because it was never going to play out like that. What you can do this quarter is make a handful of moves that pay off no matter what: moves that set the table and get you going. Here are four. Each is specific, and each is worth doing under every version of the future your company grows into.

1. Internalize that off-the-shelf AI won’t deliver the magic on its own. The first level of AI adoption comes with a subscription, and it does deliver real productivity gains, but it falls far short of the AI-powered future we’re all expected to build. Most organizations are stuck in exactly that place today, and for good reason: Off-the-shelf AI tools have PhDs in everything yet know nothing about your business. That makes them world-beaters at general tasks, but clumsy new hires for real business work. Closing this gap takes work, and no subscription does it for you. But it isn’t rocket science. It’s just a traditional software and data problem. The hardest part is simply seeing the gap, and understanding that closing it is the mission.

2. Take stock of what’s already being built, and who’s building it. Because the gap is a software and data problem, it helps to know what your in-house software talent looks like. First, does your company currently employ software developers? If so, are they using coding agents such as Claude Code, Cursor or Codex? And are they building custom AI agents yet? Then walk the floor for homegrown agents and automations already running, and notice that whoever built one might not be a developer at all. The people quietly automating their own corner of the business are sometimes exactly the builders you’re looking for.

3. Find your “if only someone were always watching this” jobs. The best “wave one” agents don’t fill roles you want to eliminate. They do jobs nobody’s doing. Look for the processes where you’ve thought, “If I had one person constantly watching this and thinking about it, things would get meaningfully better, but I’d never justify a full hire for it.” Ask managers at all levels to run this mental exercise. Then focus on the low-hanging fruit you turn up: things you could train a new hire to do within two weeks. These will be your safest, most confidence-building starting points.

4. Get started on semantic models (if you haven’t already). A semantic model is a decoder ring for your business’s data. It’s a set of blessed, machine-readable definitions that turn raw records into the terms your business actually runs on: what counts as an “active customer,” say, or which expenses do and don’t belong in “gross profit.” Without one, AI agents can’t reliably interpret the data in your systems of record. Which is why every software vendor in the data industry is now racing to be your semantic model provider. If you’ve already invested in Power BI, you may have a head start, but don’t get too hung up on which tech to adopt, because the definitions themselves are the real work. Once captured, they tend to be portable.

None of these four asks you to predict the future or bet the org chart. That’s what makes them particularly valuable in the face of today’s uncertainty. Whatever your AI future state turns out to be, and it will keep evolving, you’ll be glad you did all four.