The monthly AI review is well attended, carefully prepared, and produces almost nothing an executive can act on. What the work is missing is a rhythm that turns evidence into owned decisions.
The meeting that keeps meeting
Pull the last three agendas from your AI steering committee.
The pilot list is longer than it was. The agent inventory has grown. Someone presents an adoption dashboard. Someone raises a data access exception that first came up two months ago and is still open. The vendor roadmap gets a slide. Everyone leaves the room informed.
Now look at what actually left the room with them. Which workflow got funded for scale. Which agent got rebuilt against a tighter workflow. Which exception got closed. Which tool got stopped. Which name is attached to each of those, with a date.
For most companies the honest answer is that the meeting produced alignment and deferred the decisions. The same three questions return next month with better slides.
That is the tell. The forum has become a reporting surface for AI activity, and the decisions it exists to make keep circulating.
Deferred decisions don't stay still. They accumulate: pilots waiting on a scale call, exceptions that stay open past the meeting that first raised them, agents nobody has checked against a written standard, and vendor or ownership questions nobody has closed.
A reporting forum and a decision forum are different instruments
Steering committees are good at some things. They socialize direction, surface cross-functional conflict, keep sponsors informed, and give a program a visible executive shadow. That's real work, and it's worth doing.
They are structurally weak at deciding. Attendance rotates. Authority is diffuse. Consensus is the exit condition, so the safest outcome in any given hour is to gather more information and revisit. When nobody leaves with a decision on their name, the decision comes back.
Governance belongs to a person, not a committee is a case Anchor has made before. An operating cadence is where that principle stops being an org-chart preference and becomes a calendar.
The distinction worth drawing is between a forum that informs and a forum that decides. Only one of them changes what happens next week.
What an operating cadence actually is
An operating cadence is a recurring executive rhythm with three properties.
Named decisions sit on the agenda before the meeting starts. The agenda carries decisions phrased as decisions. "Agent performance" is a topic. "The support triage agent has missed its accuracy standard for three consecutive weeks: rebuild it against a narrower workflow, or stop it" is a decision, and it can be closed in the hour.
Evidence arrives before the meeting. The forum spends its time judging what the evidence says, and it spends none of that time assembling it.
One accountable owner is attached to every decision, with the authority to move it and the exposure if it doesn't move.
Decisions then get recorded with a verb: scale, govern (keep a working workflow running under clear controls and named ownership), rebuild, or stop. Those four verbs are how Anchor already worked through the growing pilot graveyard, and they carry cleanly into agents, workflows, and vendor commitments.
The proposed rhythm
The cadence Anchor proposes runs three loops, each tuned to the speed of the thing being decided. The intervals below are a starting design, not a standard. A business with a monthly close and a quarterly board cycle will tune them differently than one shipping weekly.
An organization early in AI adoption may start with the monthly decision forum alone and add the weekly and quarterly rhythms as its portfolio and its workflows mature enough to need them.
The weekly loop is the answer to the exception that's still open two months after it was first raised. It's operational, and it doesn't need an executive in the room. In practice, the workflow owner may be supported by a small team: someone checking agent output against the written standard, a security liaison for access exceptions, a change contact tracking adoption friction. The cadence still needs one name attached to the decision, even when the evidence comes from several people.
What gets reviewed each week: agent output sampled against its written standard, exceptions raised and how they were escalated, access requests that fell outside the approved path, and adoption friction reported by the people actually doing the work. This is the same standard-and-drift discipline that keeps an agent honest from one week to the next.
Most of what the weekly loop produces is a closed exception. What it can't close, it escalates with the evidence already attached.
The monthly loop is the portfolio decision forum, and this is where the executive sponsor belongs. It is where the pilot list from the opening diagnosis actually gets decided instead of re-presented. It reviews workflow value against baseline. It takes the agents and pilots the weekly loop has flagged and applies the four verbs to them. It closes the exceptions the weekly loop escalated, or it names who will close them and by when. This is the meeting where the AI portfolio stops growing by accretion.
The quarterly loop is the executive reset, and it exists because some drift is too slow for a monthly meeting to catch. It is the altitude where a COO, CIO, or CAIO asks the questions that don't fit inside a month. Has vendor dependency quietly become operating dependency? Is the approved path still faster than the workaround it was meant to replace? Which capability has the company decided it will not build? And is the cadence itself still earning its place on the calendar?
The evidence the cadence runs on
A cadence without evidence turns back into a status meeting within two quarters.
The evidence set has to stay small enough that a workflow owner can produce it without a reporting program behind them. Anchor proposes five standing artifacts:
- Workflow value, measured against the baseline that existed before AI touched the workflow. If that baseline was never captured, establishing it is the first finding, and the workflow's value claim stays provisional until it is.
- Agent performance, sampled against a written standard, with the difference from the last trusted run made explicit.
- An exception and escalation log showing what was raised, who closed it, and what remains open.
- An adoption signal from the people doing the work, including where the approved path is losing to the workaround.
- A vendor and dependency note covering what the company would have to rebuild if the contract ended.
Five artifacts, each owned by a name, each prepared before the forum meets. When an artifact cannot be produced at all, that absence is itself the finding, and it belongs at the top of the agenda ahead of anything else on it.
One owner, or the rhythm decays
The convener of a cadence and the owner of a decision are different roles. Conflating them is how a decision forum quietly reverts into a steering committee.
A convener schedules, distributes, and follows up. An owner controls what the decision actually requires: the budget line, the workflow, the data path, the people, and the authority to stop something a sponsor already praised in a board update. Someone holding the first set of responsibilities and none of the second is a coordinator, and a coordinator cannot close a decision. They can only carry it back to the room it came from.
That produces a practical test to run on your own cadence. For each open AI decision, name the single person accountable for moving it, and name the approvals they will actually need to clear. If the answer is a committee, or "it depends," the decision will still be open next quarter.
Designing against the failure modes
Three ways a cadence decays, all of them predictable enough to design against.
It becomes theater. Status creeps back onto the agenda, decisions slide, and the forum turns into the thing it was built to replace. The guard is the decision log. When a full cycle produces no decisions, the cadence has a problem that someone should name out loud.
It becomes a reporting burden. The evidence set grows, because every request for one more chart sounds reasonable in isolation. The guard is a hard ceiling on standing artifacts, plus the willingness to retire the ones that stopped changing anyone's mind.
It outlives its usefulness. A workflow stabilizes, the exceptions stop arriving, and the decisions get boring. That is success. The response is to retire the item from the cadence and convert the stable part into an operating control. A cadence that only accumulates is a cadence nobody will keep.
Before the next AI review
Pull those three agendas again and mark every question that appears on more than one of them.
Those repeated questions are your inventory of unowned decisions. Each one is still sitting in the room because the evidence wasn't prepared, or the question was never phrased as a decision, or nobody present had the authority to close it. The recurrence is the diagnostic, and it's available to you this week without a new tool, a new hire, or a new governance program.
An AI Bearing Assessment identifies where a leadership team's operating model is weak, a diagnosis made at the level of strategy. An operating cadence is what turns that finding into owned decisions through the next quarter.
The AI work inside your company is already producing decisions. The open question is whether anyone is making them on purpose.