|
Issue 009 · June 18, 2026
|
The Signal
A weekly read for operators thinking seriously about AI. Less noise, more signal.
|
|
|
|
Move — the committed action
A team finishes the logic. The system can now make the call. They demo it to leadership, the demo goes well, the project gets greenlit for rollout. Six weeks later the AI is still flagging recommendations and the humans are still doing the work. Nothing downstream has changed. The system has opinions; the organization has habits. The two are not connected.
That is the Move problem, and it is where most AI implementations die without anyone calling it. Making the call and acting on the call are not the same thing. A decision that does not connect to a committed action is a recommendation, and recommendations are advice. Advice scales poorly. Move is the layer that turns a decided call into a thing the system actually does — or, when that is not appropriate, the layer that hands the call to the right person in the right form at the right time.
The mistake teams make at this layer is assuming the action falls naturally out of the call. It does not. A "qualified lead" call is not an action. A "qualified lead added to the priority queue with the next-step playbook attached and the assigned rep notified within five minutes" is. The difference between those two sentences is the entire Move layer.
Three things define a Move for any decision. The act, the path, and the commitment level.
|
|
01
|
The act
The act is the specific thing that happens when the decision lands in a given bucket. Not "the lead is qualified" but "a CRM record is updated to qualified status, a Slack message is sent to the assigned rep, and a follow-up task is scheduled for the next business day with the qualification rationale attached." The act is observable, executable, and produces a state change somewhere a human or a downstream system can see.
Every lens bucket needs its own act. The contractor's three buckets — bid-fast, bid-careful, decline — produce three different acts. Bid-fast generates a one-page proposal in Google Docs, attaches it to the inquiry email thread, and notifies him via text. Bid-careful triggers a calendar block for a site visit and a draft proposal pending the visit notes. Decline sends a polite referral email to a partner contractor and logs the decline reason for monthly review. Same trigger, same frame, same logic, three completely different acts. If the buckets all produced the same downstream action, you would not need the buckets.
|
|
|
02
|
The path
The path is where the act actually lands and how it gets there. Which system holds the resulting state. Which API or integration produces the change. Which human, if any, is in the loop. This is the layer that sounds like a tooling conversation but is really a workflow conversation. The wrong path makes a well-designed act invisible. A perfectly framed lead qualification that updates a field nobody looks at is not a Move; it is paperwork.
The fastest way to choose the right path is to ask where the result of the action is already being checked. If your sales team lives in the CRM, the path lands in the CRM. If they live in Slack, the path lands in Slack with a CRM mirror. The path should run through the surfaces the team already pays attention to, not the ones the implementation team finds elegant. Elegant paths that nobody watches are the most common failure mode in this layer, and they are almost always the result of treating Move as a technical decision rather than a behavioral one.
|
|
|
03
|
The commitment level
Commitment level is how far the system goes before a human gets a vote. There are roughly four positions on the spectrum and you have to pick one explicitly for each act. Recommend — the system suggests; a human acts. Stage — the system prepares the action and a human releases it. Execute with audit — the system acts and a human can reverse it within a window. Execute autonomously — the system acts and the human only sees the result.
Most teams set commitment level by accident rather than by design. They default to Recommend because it feels safe, and then wonder why six months later nothing has changed downstream. Recommend is the safest position and the least valuable; the system does work, but the work that follows depends entirely on whether the human acts on the recommendation, and in busy operations the answer is usually no. Stage is the position most teams should be picking and almost none do. The system writes the email, drafts the proposal, fills in the CRM update — and the human just hits send. That hand-off is where the productivity gain actually lives. The commitment level should be chosen for each act independently, based on the consequence of being wrong and the speed cost of waiting for a human.
|
|
Act, path, commitment level. Three sentences per bucket. Most teams can write them in an afternoon. Most teams do not, and the cost of skipping shows up as the silent stall — the six-month version where the system is technically deployed and operationally invisible.
One honest moment, because this layer is where the politics show up. Commitment level is not just a technical setting; it is a question about how much your organization trusts the system you built. The conversation about whether the AI should "send the email" or "draft the email for review" is rarely an engineering conversation. It is a conversation about who is accountable when the email is wrong, how much the team has internalized the logic the system is using, and whether the operator at the receiving end of the path believes the system is operating from their lens or from someone else's. Move is the layer where Decision Architecture stops being a methodology and starts being organizational change.
This is also why so many AI initiatives stall here without anyone noticing. The technical work is done. The system makes good calls. The calls are just not connected to anything that matters. Move is the layer that closes that loop. Skip it and the rest of the method is theatre.
|
|
Free Worksheet · Part 5
The Move worksheet
Three sections — the act per lens bucket, the path it lands on, the commitment level you pick — designed for one decision from your audit. Reply with "move" and I will send it directly. Missed earlier worksheets? Reply "all" and I will send the full set so far — Audit, Trigger, Frame, Logic, and Move.
|
|
|
Next Week · Part 6 of 6 · Series Finale
Layer 06 — Echo
The final layer. Every encoded decision leaves something behind — a record, a signal, a small piece of organizational memory. Echo is the layer that turns those traces into the system that learns. Without it, decision architecture is a snapshot. With it, it is a flywheel.
|
|
If you are looking at one of your own decisions and finding that the AI you bought is stuck at Recommend when it should be Staging or Executing — and the team would be fine with that change but nobody is making it — that is the conversation a Discovery call is for. Bring the act, the path, and the current commitment level. We will work through where it needs to move.
Or read this issue on LinkedIn.
|
|
|
|