You Don’t Need to Learn AI. You Need to Lead It.

A lot of executives worry they’ve fallen behind on AI, and many have the same plan to fix it: take a course, learn to write better prompts, maybe get a certificate. I understand the instinct. I also think it’s mostly the wrong use of a senior leader’s time.

You don’t need to become an AI practitioner. You need to lead an organization that’s going to use AI whether you direct it or not. Those are different jobs, and the second one is yours.

Prompting was the tutorial level

For the last couple of years, “using AI” has mostly meant one person, one chat window, one question at a time. Draft this email. Summarize this document. Rewrite this paragraph so it sounds less angry. That’s useful, and if your people aren’t doing it yet, they should be.

But it’s the tutorial level. The value from a chat window is capped by how many minutes a person spends typing into it. It makes individuals a bit faster. It doesn’t change how work moves through your organization.

The next stage is already here, and it looks different. AI agents and automated workflows can take a task that crosses several systems, carry it through, and hand a person the result to approve. In my own shop I run an agent that checks the health of my cloud servers on a schedule and handles routine fixes before I’d otherwise notice them. I have publishing workflows that check their own output and roll themselves back when something doesn’t match. None of that came from better prompting. It came from deciding which work should be done differently, then redesigning the process around it.

That redesign is a leadership decision. An engineer can build the workflow. Only someone with authority over the process can decide that the process should change.

Why this is a leadership problem

I spent thirty years in the Army, a good part of it bringing new communications technology into organizations that had to keep operating while we did it. The technology was rarely the hard part. The hard parts were the same every time: getting people to trust the new system, deciding who owned it, changing the procedures that assumed the old way, and making sure the mission didn’t stop while all of that happened.

AI adoption follows the same pattern, with a twist. Your people are already using it. Some are using it well. Some are pasting customer data into free tools you’ve never vetted. A few are quietly doing two hours of work in twenty minutes and wondering whether to tell you, because they’re not sure if they’ll be rewarded or handed more work. If you don’t set direction, you still get AI adoption. You just get the uncoordinated, invisible version, with the risk and without most of the benefit.

That’s why I think the organizations that embrace AI deliberately are likely to pull ahead of the ones that don’t. Not because the technology is magic, but because they’ll learn faster. Every workflow they automate teaches them something about the next one. A company that spends a year experimenting with intent will be in a very different place from one that spent the year waiting for a perfect policy.

What leading AI actually looks like

You don’t need to know how a transformer works. You do need to do these things, and they’re things nobody below you can do on your behalf.

Pick the work, not the tool

The most common mistake is starting with a product. Someone buys licenses, rolls them out to everybody, and waits for value to appear. Usually it doesn’t, or it shows up as scattered personal productivity nobody can measure.

Start with the work instead. Which processes in your business are repetitive, rules-based, and painful? Where do people copy data from one system into another? Where does a request sit in a queue for three days waiting for someone to look at it for three minutes? Those are your candidates. Pick two or three, define what good looks like, and point your effort there.

Set the guardrails out loud

People need to know what’s allowed. Which tools are approved? What data can go into them? What has to be reviewed by a person before it goes out the door? For defense contractors, this isn’t optional. Controlled information in the wrong AI tool is a compliance incident, and I’ve written about exactly how that happens in Is Your Prompt History a CUI Spill?

The guardrails don’t have to be perfect on day one. They do have to exist, and they have to be written in plain language. “Don’t use AI for anything sensitive” isn’t a policy. “Use the approved tool in our tenant for internal documents, never paste in customer drawings, and have a person review anything that goes to a customer” is.

Make it safe to talk about

If your best people are hiding how they use AI, you have a culture problem, and it’s costing you. Ask openly. Have people show what’s working in team meetings. Reward the person who automated half their week by giving them a harder, more interesting problem, not by quietly assigning them twice the volume.

The flip side matters as much. When someone makes a mistake with an AI tool, like pasting the wrong file or trusting a confident wrong answer, they need to know that reporting it right away is the correct move. Punish the cover-up, not the error.

Keep a person accountable

Every automated workflow needs a human owner who’s responsible for what it does. AI can draft, sort, summarize, check and route. It shouldn’t be the last word on anything that moves money, signs a commitment, or speaks for the company to a customer. “The AI did it” won’t satisfy a customer, a contracting officer, or a court, and it shouldn’t satisfy you.

Measure something real

Hours saved is a weak metric because nobody tracks it honestly. Better ones: cycle time on a specific process, error rates before and after, backlog size, how fast you respond to a customer. Pick a number before you start so you can tell whether the change worked.

Three mistakes I’d avoid

A few missteps come up again and again, and they’re all avoidable.

Banning it. Some organizations respond to the risk by blocking every AI tool. It feels responsible, and it doesn’t work. People use their phones and personal accounts instead, which moves the activity somewhere you can’t see it at all. You end up with the same data exposure plus zero visibility. Pick an approved tool and make it easy to use instead.

Buying licenses for everyone and calling it a strategy. A seat for every employee with no defined use cases produces a burst of enthusiasm, then a plateau. Six months later, finance asks what the spend is buying and nobody can answer. Licenses follow use cases, not the other way around.

Handing it all to IT. Your IT team should own the platform, the security controls and the integrations. They shouldn’t be the ones deciding how your sales process, your proposal shop or your contracts team ought to work. That’s an operations decision, and it belongs to the leaders who own those operations. When IT owns AI alone, you get well-secured tools that nobody’s process was redesigned to use.

The workforce question

Sooner or later every leader asks some version of “what happens to my people?”

My honest answer is that roles will change, and some tasks people do today will mostly go away. The organizations that handle this well will be the ones that treat their current staff as the people who’ll run the new processes, not the people who’ll be replaced by them. Your employees know where the friction is. They know which steps are pointless, which exceptions come up every week, and which customer always needs special handling. That knowledge is exactly what you need to automate well, and you lose it if people think helping will cost them their jobs.

Training matters here, but it’s not the same course you’d take. Your workforce needs practical skills: how to use the approved tools, how to check AI output, how to describe a process clearly enough that it can be automated. Your managers need to learn how to run a team where some of the work is done by software that needs supervising. That’s a management skill, and most managers haven’t been taught it yet.

I’ve written separately about how agents change workforce economics and about why agents need a strategy rather than a shopping list.

Where to start this quarter

If you want something concrete, here’s what I’d do in the next ninety days.

• Name one executive owner for AI adoption, someone with authority to change processes, not just buy software.
• Publish a one-page acceptable use policy in plain English.
• Pick two or three processes to redesign, with a measurable before-and-after for each.
• Find the people already using AI well and make them part of the effort, visibly.
• Review results at the end of the quarter, keep what worked, and pick the next processes.

None of that requires you to learn to write a better prompt. It requires you to do what leaders have always done when a new capability shows up: decide where it fits, set the rules, bring your people with you, and keep the mission running while it all changes.

---

Where to go next

• Defense contractor planning AI adoption? Check your baseline first with the CMMC readiness quiz.
• Our CMMC guide for small contractors walks through scoping from the beginning.
• Need the SSP and policy templates underneath an AI rollout? Start with our CMMC packets.