AI-First Is Not a Tool Decision — It’s an Operating Model Decision

Every hand goes up when I ask who uses AI tools. Every hand goes down when I ask who has an AI strategy. I’ve run this test in enough workshops now that it no longer surprises me — but it still bothers me.

Because in that gap, right between enthusiastic adoption and deliberate design, that’s where the money quietly disappears.

The Real Problem Isn’t the Tool. It’s the Thinking Behind It.

This week I was at an event in Freiburg — a summer gathering of pharma and med-tech folks — and I found myself explaining Pheraxia for the first time in public. The science landed quickly. People got it: take bioactives that already work and make sure they actually reach the cells they’re supposed to help. Obvious problem. Obvious need.

Then came the question that stopped me.

“So you’ll need accounting staff, marketing, people to build the simulation models — and you want capital to hire them first, then run lab tests?”

It was asked genuinely. And it told me everything about the real challenge ahead — not in pharma, but in every industry trying to go AI-first.

The mental model of company = headcount = capability sits so deep that an asset-light, AI-operated structure doesn’t read as innovative. It reads as incomplete. As if we’d simply forgotten to hire people.

We hadn’t forgotten. We’d made a different choice. AI does the heavy lifting — design, simulation, operations. The team stays deliberately lean. Capital flows toward proof, not payroll.

That’s not a smaller company. It’s a different kind of company.

Why Tool-First Thinking Fails (And What to Do Instead)

Back in the workshops: the pattern I keep seeing in Mittelstand companies isn’t laziness or lack of ambition. It’s sequencing error.

Licenses get bought. A prompting workshop happens. Three months later, either nobody uses the tools consistently — or, worse, everybody uses them individually and leadership has no visibility whatsoever.

Both outcomes are expensive. One wastes the license fee. The other creates invisible risk.

A colleague I respect put it cleanly in a recent conversation: if your first AI step is a tool rollout, you’ve probably already taken a wrong turn.

I’d add: the wrong turn isn’t buying the tool. It’s buying the tool before you understand the process it’s supposed to improve.

This is the exact same mistake I’ve watched derail transformation projects for fifteen years — in operations, in post-merger integration, in finance. The technology was never the hard part. The thinking about what the technology is supposed to do, and who decides when it’s done right — that’s always been the hard part.

The Distinction That Changes Everything

Here’s a frame I keep returning to, and I want to name it clearly:

Topics vs. Processes — The AI Readiness Test

A topic is a direction. “We’re going to use AI in sales.” Fine. But a direction isn’t executable. You can’t automate a direction. You can’t govern it. You can’t improve it.

A process is a specific, repeatable sequence with defined inputs, decisions, outputs, approvals, and feedback loops. “Every inquiry from the contact form gets a vetted draft within one hour, reviewed and approved by a human before it goes out.” That’s executable. That’s automatable. That’s improvable.

You cannot delegate topics. You can only delegate — or automate — processes.

This isn’t a subtle distinction. It’s the one that separates companies building real AI capability from companies accumulating AI subscriptions.

The Five-Field Process Card

When I work with CEOs and operators on AI readiness, I use a simple diagnostic before anyone touches a tool. I call it the Five-Field Process Card — it fits on an index card, which is the point.

Take one recurring process in your business and answer these five questions:

What comes in? (What is the trigger, the input data, the starting condition?)
What needs to be decided? (Where is the judgment call? What makes this non-trivial?)
What should come out? (What is a good output — and how do we know?)
Who approves? (Who is accountable for the decision before it leaves the building?)
What gets better after each run? (What do we learn, log, or refine so iteration compounds?)

If you can’t answer all five, you don’t have a process yet. You have a habit — and habits don’t scale, don’t transfer, and don’t survive leadership transitions.

Once you can answer all five, you’re ready to ask: which of these steps can AI handle, and which require human judgment?

That’s the right order. Process clarity first. Tool selection second.

”Completely Thought, Leanly Built”

One more tension worth naming. In a recent episode of 10xCEO, we debated whether AI adoption should start small and grow, or be designed holistically from the start.

The honest answer: both concerns are valid.

Start too small without systemic thinking, and you build isolated automation — faster, but not better. Individual pockets of efficiency that don’t connect, don’t transfer, and create shadow workflows leadership can’t see.

Design too comprehensively without building anything, and you’re back to strategy theatre.

The working compromise I’ve landed on: completely thought, leanly built. Design the full logic — inputs, decisions, outputs, governance, learning loops — before you build. Then build the smallest possible version that tests whether the logic is real.

This is how Pheraxia is built. AI as the operating layer. Humans governing the decisions that matter. No lab, no large team, no traditional overhead. But the architecture — the thinking — is complete.

That’s not a startup shortcut. That’s a deliberate operating model.

What You Can Do Next Week

If you’re an SME owner, CEO, or operator reading this, here are five concrete moves:

Run the hand-test yourself. Ask your leadership team: who uses AI tools personally? Then ask: do we have a written AI process anywhere in the company? The gap you find is your starting point.
Pick one recurring process — just one. Not a topic. A specific, repeatable sequence that happens at least weekly. Sales follow-up. Reporting preparation. Supplier communication. One.
Fill out the Five-Field Process Card. Write down what comes in, what’s decided, what comes out, who approves, and what improves. If you can’t complete it in 20 minutes, the process isn’t defined enough to automate.
Separate tool conversations from process conversations. In your next leadership or ops meeting, explicitly ban tool names for the first half. Talk only about what you want to achieve and how decisions currently get made.
Name one human checkpoint. For any process you’re considering automating, decide now who reviews the output before it reaches a customer, partner, or regulator. Governance isn’t bureaucracy — it’s what keeps AI-first from becoming AI-reckless.

The hardest thing I’ve had to explain this week isn’t the science behind better bioavailability, or the logic of an asset-light structure. It’s that doing less with AI isn’t a failure of ambition — it’s often a sign of clearer thinking.

AI-first doesn’t mean AI-everywhere. It means AI where the process is ready, the decision rights are clear, and the human in the loop knows exactly what they’re governing.

Everything else is a subscription you’ll quietly cancel in six months.

Which process in your company would you put on the Five-Field Card first — and what stops you from starting there?

AI-First Is Not a Tool Decision — It's an Operating Model Decision