Before Phase 1: name the people who own the initiative

Here's a question I like to ask early: who in your company owns AI? I don't mean who bought the licenses or who gave the kickoff talk. I mean who decides which phase you're actually in, what runs next, and what never ships.

If the answer is "sort of everyone," that's the gap this post is about.

Our AI Transformation Playbook has four phases: Fluency, Gated Agents, Agentic Loops, and Learning Loops. But step one isn't Phase 1. It's a prerequisite: stand up an AI project / leadership team, and give it the time and the mandate to steer. Fluency without a steering body tends to produce pockets of experimentation, a few enthusiastic demos, and nobody who owns the initiative.

A title is not a team

The market has noticed that AI needs an owner. IBM's 2026 CEO study (2,000 CEOs and senior leaders, surveyed February to April 2026) reports that 76% of surveyed organizations have a Chief AI Officer, up from 26% a year earlier. BCG's AI Radar 2026 (nearly 2,400 executives, including 640 CEOs) found that nearly three-quarters of CEOs say they are their company's main decision maker on AI.

That's progress, but an owner at the top isn't people with time to do the work. The same IBM study reports that 83% of surveyed CEOs say AI success depends more on people's adoption than on technology. Adoption happens in departments, on real tickets. A title doesn't get there by itself. A team does.

Who sits on it

Three roles, all named:

• An executive sponsor. One leader who can clear obstacles, settle priority fights between functions, and answer for the work at the leadership table. One person. Not a committee, and not "IT."
• A chair. The person accountable for the team week to week. That might be a dedicated AI or transformation lead, or the chair of a cross-functional working group. Either works, as long as one person holds it.
• A small team. The playbook suggests these seats: someone who knows your operations, someone who knows your systems and data, someone who owns risk and review, and a rotating seat for whichever function is adopting right now.

Keep it small: a working group, not a committee that meets quarterly to approve slides.

The mandate fits on one page

Write it down. The playbook gives the team five jobs:

1. Locate the phase, honestly and per function where it differs, and keep that picture current.
2. Sequence the plays: what runs now, what waits, and what never ships.
3. Protect the operating ceilings. Some outputs, like anything that reaches a child or family, stay human-gated permanently. Someone has to own those calls in writing.
4. Decide what ships externally and what stays internal, and who signs off on each external channel.
5. Hold teams accountable for exit artifacts, not attendance at workshops.

Protected time, or it didn't happen

This is the part that gets negotiated away. The team has two kinds of work. The first is building and maintaining the roadmap: phase location, sequencing, the candidate list, ceiling decisions, and an honest status read for leadership. The second is working inside departments so AI takes hold in real workflows: sitting with teams, running sessions on their actual emails and docs, unblocking approvals, and following up. Buying licenses isn't adoption.

Dedicated people are best. If that's not possible, members need substantial protected time: real calendar hours their managers have agreed to protect. A practical starting shape: a standing weekly working session for the chair and core members, department time in between, and a monthly sponsor review of the roadmap, artifacts, and open decisions.

My test is simple. If the weekly session keeps getting cancelled, the team doesn't have the time it was promised. Fix the allocation before you blame the plan.

What exists before Phase 1 starts

The exit artifacts for this step are deliberately boring:

• A named executive sponsor
• A named team and a named chair
• A one-page written mandate
• A time allocation note: who, how much protected time, agreed with their managers
• A first 90-day roadmap stub: phase located, first plays sequenced, review cadence set
• An interface list of department partners: the function lead and a fluency owner for each team

That last one matters more than it looks. The fluency owner in each team is the hand-off point for Phase 1, the person who can answer "what are we trying, what's approved, and what goes external?" The team partners with function leads instead of routing around them, and it writes its decisions down where departments can see them.

None of this needs a vendor. It needs a few decisions and a couple of signatures.

The usual failure

Skip this step and the failure tends to follow a familiar pattern.

It starts with someone being told to "own AI" on top of a full job. The playbook calls this AI as a side hustle: the team exists on paper, nobody has protected time, and the working sessions slide off the calendar one by one. Meanwhile there's a kickoff keynote, maybe a town hall, and real energy in the room. Then nothing comes out the other side. That one is Podium theater: events, no artifacts.

Underneath, there's often a third problem: no sponsor. When two functions want the same people or data, the fight stalls and nobody answers for the work. The mirror image is the central team that never visits. The roadmap looks clean, and the departments haven't changed a thing.

Microsoft's agentic AI adoption maturity model describes the starting point plainly. At Level 100, "Initial," the strategy pillar lists "no executive sponsor", and an executive sponsor first appears at Level 300, "Defined." I don't read that as a score to chase, just a reminder that ownership is foundation, not finish.

The repairs are just as plain:

• No sponsor: name one executive and put AI on their monthly agenda.
• Side hustle: write the time allocation note and get managers to sign off.
• Never visits: work in departments every week, co-owned with function leads.
• Podium theater: judge the team by its exit artifacts, not its events.

What to do this week

A Monday-sized version: pick the sponsor. Pick the chair. Ask the chair to draft the one-page mandate and the time allocation note, and get the relevant managers to sign the note. Book the first weekly session and the first monthly review.

Already running agents with nobody owning the initiative? Don't tear down what works. Stand up the team now and have it locate you honestly. Between two phases, pick the lower one.

The team doesn't dissolve when Phase 1 starts. Its job shifts from setting up to steering, through every phase after.

Get the playbook

The full prerequisite section, failure-modes table included, is in the 231 Labs AI Transformation Playbook, along with the four phases that follow it. It's free at 231labs.com/playbook.