To get there, you'll need a map

James wrote last week about tuning out the AI noise and getting stuff done. And there's a lot of noise. The hype, the doom, the new model every Tuesday. One line of his stuck with me. "Where people get into trouble," he wrote, "is when keeping up with AI starts taking the place of the work."

He shared the story of our client who had nearly every agent out there running on his own Mac mini. Impressive setup. But when we asked what problem it was solving for his business, he didn't know how to answer. He was going 95 mph on a highway to Who-Knows-Where.

So it's worth stopping for a second to tune out the noise. Once it's quiet, take the time to decide where to go.

AI can write the code and hand you a dozen options before your coffee's done. But it can't tell you what your priorities are. Those are yours to define, and that takes knowing how your business actually runs.

This is a point so many guests on DTAScience keep landing on, no matter what field they come from.^[1]

Dr. Tshaka Cunningham uses AI tools all the time. "But I don't use AI to generate my idea," he told me. "That comes from deep within." His rule is that your gut, your "seed of an idea," "should always be above the AI layer."

Tommy Tang uses Claude Code every day for bioinformatics, and he still says the hard part hasn't moved. "I think the hard part of a good bioinformatician is first to frame the questions well."

David Aanensen, who's spent his career on global pathogen surveillance, came at it through ownership. The question he cares about is "who owns the capability to detect, interpret, and act on their own data."

Dr. Luke Short runs the Dallas County Public Health Laboratory and uses AI as a second set of eyes on lab procedures. But when it flags something, he takes it to the expert, and the expert makes the call.

Dr. Sandeep Pawar has watched pharma AI pilots shine on clean data and stall at scale. His advice is to go "methodically and thoughtfully" and stop trying to "spray and pray." Point AI at the wrong step and you can optimize it to the nth degree and still end up right where you started. Fastest car on the lot, wrong exit.

Different fields, same landing spot. Someone who understands the work has to pick where "there" is, then sit down and map it out, on purpose, before anything gets built.

AI makes that step easy to skip. You can go from idea to prototype in an afternoon, so it's tempting to floor it and figure out the route later. Most teams I meet start from "we want to use AI." But that's the car, not the destination.

Every map starts with a little dot that tells you where you are. For a team, that's an honest picture of your process, where things are working and where they're stalling. Knowing this is critical before you hit the gas on all things AI.

Brandon Galang, an industrial and systems engineer by training, starts with "What is the bottleneck?" AI can even help you "map the system and then even generate potential hypotheses" to test. But you still decide what's worth mapping.

We call this building a Friction Map. It's how you put that dot on the page. Pick the repeated tasks your team grumbles about, the ones keeping the business from moving forward. The map turns your estimates into a picture of the weekly hours and yearly staff cost tied up in each task, then flags what's ready for AI and what needs work first.

That's your starting point, not a promise of savings. Treat those hours as your first estimate, then track the task for a week before you change it. Now you've got an odometer reading you can actually measure progress against.

We've learned this lesson ourselves. Some of our public case studies still label business KPI impact "Not instrumented." A client-reported improvement is useful, but it isn't the same as a measured before and after.

But knowing where you are doesn't tell you where you're going.

Once you know where you are, look up at where you want to be. Maybe it's getting reports out the door a day faster. Maybe it's never letting a lead go cold. Write it down in plain language, with what a good result looks like and how you'll measure it. (Back in bioinformatics, we called that a design doc, and the best projects I worked on never started without one.)

Then go back to your map and plot the route. Some tasks look like a great fit for AI until you look closer. If a task runs differently every time, the map says fix the process first. Automating an inconsistent process can just make the inconsistency happen faster. If the information lives on paper, comes in by phone or sits in someone's head, it says get it into a system first. And if a mistake would put money, safety or compliance on the line, it keeps a person in the driver's seat.

We recently scoped a roadmap for a small nonprofit that had just moved its program records off paper and into shared drives, with data scattered across forms and spreadsheets. So the roadmap starts with the unglamorous stuff, like one system of record per program, a standard way to file documents and a repeatable way to pull numbers for grant reports. AI-assisted data entry comes after that groundwork is in place.

The Friction Map shows you the terrain and flags what needs attention first. You still choose the destination. Start with one task, define the result you want and decide who checks the output. If you want help plotting the route, our Friction Audit builds on the map with your team to prioritize the work and scope the first build. Then compare the result with what you measured before.

That attention to the work matters. ColoWatch reported complex marketing production going from days to hours. And Brambleton Community Association's general manager said we "took the time to understand what we were looking for before proposing anything." That's this whole article in one sentence.

That's a reason to hit the gas, not just a chance to go faster.

We built a tool to help you build your Friction Map. And it's free.

Pick one or two repeated tasks (lead follow-ups, intake forms, reports, invoicing, the usual suspects) and answer five questions about each. Nothing connects to your systems. A one-page map lands in your inbox, and the rules behind every call are published right on the page.

Start with the task you already wish someone would take off your plate. If the map says it's ready, you've got a place to begin. If it says fix the process first, you've still learned something useful before spending money on a tool.

AI is going to keep getting faster, and there'll always be a shinier whip. A good map is what points all that speed where it matters for your business.

Get your Friction Map →