The commitment nobody made on purpose

A CEO I’m working with, call her a $5M services operator, fifty people, owner who’s also the CFO and de facto head of HR, has spent somewhere between $4,000 and $8,000 a month on AI tooling over the last fourteen months. ChatGPT Teams. A Claude trial that converted to seats. A Copilot pilot. Two vertical SaaS tools that quietly added AI features to her scheduling stack. An automated workflow her ops manager set up to triage inbound customer requests.

None of it is policy-governed. None of it is in writing. None of these decisions went through anything resembling a stage gate. They happened the way most things happen at this scale, somebody needed something, somebody clicked subscribe, the credit card auto-renewed, and the team adapted.

Then I asked her a question. “If a regulator or your insurance carrier asks next year for documentation of how AI is being used in this business, can you produce it?”

The room went quiet.

This is the pattern I’m seeing across my advisory book. And it’s the opposite of the AI fog problem the Fortune 500 is wrestling with right now. Toby Stuart, faculty director at Berkeley Haas, wrote a piece in HBR last week arguing that AI has made the future opaque enough that long-duration commitments, capex lines, ten-year forecasts, expensive degrees, are getting harder to justify. That’s a real problem.

It’s not the SMB problem.

The SMB problem isn’t the inability to commit. It’s the inability not to commit. It’s irreversible commitment that nobody made on purpose.

The licenses auto-renew. The team experiments without guardrails. The exposure compounds invisibly. By the time the CEO sees the bill, the dependency is real, the data has moved, and the policy is six months of “we should write that down” away from existing.

That isn’t a strategic posture. That’s drift, dressed up as adoption.

The escape hatch isn’t more tools. It’s a function.

Stuart’s prescription for the Fortune 500 is to put a small in-house team on full-time AI sensing, monitor frontier capabilities, translate them into management decisions, ask the uncomfortable questions early. He’s right. He’s also writing for companies with the balance sheet to staff that team.

You can’t.

But the function still has to exist. So the question becomes the one Stuart didn’t write: what does AI sensing look like at $5M, when there’s no team to put it in?

It looks like fractional governance. Thirty minutes a week with someone whose full-time job is translating what’s actually changing in the AI market into decisions you can make on Monday. A monthly review with three questions: what did we adopt, what got harder, what changed in the market. A written policy that fits on one page. A use-case map that gets revisited every quarter, not every five years.

Call it a fractional CAIO. Call it advisory. Call it whatever lands at your scale. The structure is the point. It’s the version of Stuart’s sensing team that pencils for the operator who is also the CFO, also the General Counsel, also the chief technologist by default.

Three principles for operating in the fog at SMB scale.

The function is non-negotiable. The headcount is. You don’t need a hire. You need a posture and a partner.

Staged commitments matter more at your scale, not less. Stuart’s best line in the HBR piece, “what is the smallest commitment we can make now that buys us information and the right, but not the obligation, to follow on with more capital?” is the entire rationale for AI governance at this scale. Buying ten Claude licenses without a use-case map and a written policy is a $5,000-per-month commitment with no exit ramp. That’s not adoption. That’s drift.

Optionality is a posture, not a project. You don’t need to restructure. You need to stop locking in. Quarterly contracts where annual ones used to be the default. Pilot use cases before department-wide rollouts. Written policies that can be amended in a Tuesday meeting, not a board cycle. The fog rewards the operator who can change direction in thirty days.

Take Action

If you’re a CEO running into the licenses-with-no-playbook pattern, that’s the conversation I have most weeks. The Perimeter Scan is the smallest commitment that buys you information. a written governance posture you can act on Monday. The Fractional CAIO retainer is for operators who want the sensing function running every week without hiring for it.

The fog is here. The skyscrapers don’t pencil. The licenses you didn’t decide to buy are already on the books.

Build for optionality. Make the next commitment on purpose.

— Kirk

Further reading: Toby Stuart, ”The Future Is Shrouded in an AI Fog,” Harvard Business Review, April 2026.

Kirk Tramble is the founder of Kizata, an AI governance advisory for privately-held CEOs running $1M–$25M companies. Berkeley EECS, Tuck MBA. Find the practice at kizata.ai.