How to Actually Put AI to Work (It's Not What You Think)

I was sitting with a business development executive a few weeks ago. Sharp guy. Spends his days building relationships with potential partners, getting on calls, opening up discussions that could turn into real deals. If you asked him what his job looked like, he’d tell you most of his work is reactive. Calls come in. Meetings get scheduled. He responds.

But that’s not what I saw when we started digging.

The Hidden Routine

Before every call, this guy had a routine. He’d research the company. He’d look at their leadership team. He’d study whatever frameworks or methodologies they’d published.

And then he’d do something I found fascinating. He’d try to build a bridge between what his company does and what the partner company does. He’d map their language to his language, find the overlaps, identify where things connected and where they didn’t.

It was thoughtful work. The kind of work that requires judgment and experience. And he was doing it 10 to 15 times a week.

Each session took about an hour. Sometimes more if the company was complex or the partnership opportunity was nuanced. That’s 10 to 15 hours a week, just on call prep. Not the calls themselves. The prep.

What the Recording Revealed

Here’s where it gets interesting. When we recorded him doing this work, talking out loud while he went through his process, something became visible that he couldn’t see on his own.

There were patterns. Clear, repeatable patterns in how he gathered information, how he organized it, how he compared frameworks, how he identified the bridge points between two companies. He thought he was doing something different every time. He wasn’t. About 80% of the process was the same. The remaining 20% was where his real expertise showed up.

We gave those recordings to Claude. Asked it what it noticed. Asked it what patterns it saw. Asked it if it had questions about what he did or why.

And wouldn’t you know it, Claude was perfectly able to help with this. Not replace the judgment calls. Not eliminate the relationship instincts. But the research? The comparison mapping? The structured prep work? Claude could do that.

So now, instead of spending an hour before each call throughout the week, he starts Monday morning and has Claude pull together briefing notes for every call that week. All the research, perfectly laid out. In less than an hour. For all of them.

He still reviews everything. He still adds his own thinking. But the 10 to 15 hours of grinding through research dropped to maybe two or three hours of reviewing and refining. That’s not a small improvement. That’s a different week.

Where Most People Get Stuck

Most people I talk to about AI are stuck in one of two places. They’re either asking it to write emails that sound like a robot learned English from a corporate handbook. Or they’re convinced it can’t help with their “real work” because their real work requires expertise.

Both positions are wrong.

The email thing is a waste of AI. And the expertise objection misunderstands what AI is actually good at. It’s not good at replacing your judgment. It’s very good at handling the structured, repetitive parts of your process that eat up your time before you ever get to apply that judgment.

But here’s the problem. You can’t hand AI the structured parts of your work until you know what those parts are. And most of us don’t. We experience our work as one continuous flow of thinking and doing. We don’t see the system underneath it.

That’s why the approach matters more than the tool.

The Five Steps That Actually Work

Step 1: Do a time audit. Not the interruptions and not the meetings. Look at where you spend time doing your real work. The stuff where your brain and experience are fully engaged. Note those efforts. Rank them by how much time they take or how often you do them.

Step 2: Pick one and record yourself doing it. Talk out loud while you work. Yes, it feels ridiculous. Do it anyway. We are terrible at abstraction. Most people don’t think in systems. They think in tasks.

But if you record yourself doing the same kind of work two or three times while narrating what you’re doing and why, you’ve just created something incredibly valuable. You’ve created data that an AI can evaluate for patterns. It will find what’s systematic in your process. It will also find the exceptions, the judgment calls, the places where you deviate from the pattern because the situation demands it.

The key is the “why.” Don’t just narrate what you’re clicking on or what you’re reading. Explain why you’re making choices. Why this source and not that one. Why you organized it this way. Why you flagged that particular thing as important. The reasoning is where the real intelligence lives.

Step 3: Give it to an AI and ask what it notices. Use Claude with a paid account. Twenty dollars a month. Ask it what patterns it sees in your work. Ask it if it has questions about what you did or why.

This conversation matters. You’re not handing off a task. You’re teaching an AI how you think about a category of work.

Step 4: Ask it how it could help you automate or speed this up. I know what you’re thinking. This is the work you bring real expertise to. You don’t want to eliminate your own job. That fear is real, and it’s also mostly unfounded.

What comes back from that conversation will have assumptions baked in that are wrong. Good. Correct them. That correction process is where you’re training the AI to understand the boundaries of what it can and can’t do for you.

Step 5: Turn it into a repeatable routine. Look at what it suggests and ask yourself honestly whether it could speed you up. If the answer is yes, and it usually is for at least some portion of the work, consider making it permanent. Claude can build what it calls a Skill, which is essentially a process it can run for you on a regular basis. Same steps, same structure, same quality, every time.

The Part That Actually Required Him

The business development executive I mentioned didn’t stop doing his job. He didn’t hand his relationships over to software. What he did was stop spending the majority of his prep time on work that was underneath his expertise level.

The research, the comparison mapping, the structured organization of information. That was work he’d been doing by hand because he’d never separated it from the judgment work that sat on top of it.

Once he could see the system, the AI could run the system. And he could spend his time on the part that actually required him.

That’s the real opportunity with AI. Not writing your emails. Not replacing your thinking. Finding the hidden system inside your expertise, and letting something else run it so you can focus on the 20% that only you can do.