Diagnostic Briefing

The Diagnostic Briefing is the paid close of the interview: within ten days of it, a facilitated half-day working session for the group that controls the budget — the sponsor, the budget owner, the platform or data lead, and, we mean this, the skeptic. Bring the person who thinks the agent program is a waste of money. The session is built to survive them.

The session starts from the interview map — the schemas you have, the connectors your agents need, the disconnects — and turns it into written findings. It is also where the product first appears, on purpose and only once: with your systems on the wall we draw where one governed Core Model would sit among them — which platforms would feed it, which agents would read from it, and which disconnects it would close. That sketch is what the Core Model Blueprint later measures.

Most agent programs stall in the gap between an impressive demo and an unreliable rollout, and the funding conversation stalls with them. Engineering says the model needs more context. Finance sees a bill that doubled while output did not. Nobody in the room can say precisely where the failure lives, so the discussion defaults to vendor narrative on one side and blanket doubt on the other.

The briefing replaces that with a shared diagnostic frame. We walk the seven stages of the context supply chain — sources, structure, semantics, validation, retrieval, agent operations, learning loop — against your actual estate, name the stage where context most plausibly breaks for you, and place your organization on the L0–L5 maturity scale with the reasoning shown in the room, not handed down afterward. It ends in writing: the findings, and a buy or no-buy on the next two rungs — the Agent Context Cost & Failure Audit and the Core Model Blueprint.

• The team that took the interview and wants the map turned into a decision within the ten days, not into a longer conversation
• The executive sponsor who believes in the program but cannot yet defend it in a budget meeting
• The CFO or budget owner who has seen the model API invoice and wants to know what it is buying
• The data or platform leader tired of every agent failure being blamed on their systems, without evidence
• Any group where the funding decision is contested and a vendor deck would make it worse

The failure pattern, in plain terms: demos run on curated context; production runs on your estate as it actually is. We introduce the seven stages using examples from your own systems, taken from the interview map.

Stage by stage, against your estate, with the people who own each stage answering in the room. Where does content live and who keeps it current? Does “customer” mean one thing across your systems, or four? Does anything check content before an agent reads it? When retrieval returns the wrong material, does anyone find out?

Your L0–L5 maturity position, derived in front of the group with the evidence on the table. This is a directional placement from structured discussion — the Blueprint measures; the briefing orients. We say which is which.

• The six levels and their cues are published on the Context Supply Chain page. — L0 to L5, each with its ‘You are here if’ cue

What the next rung would cost, what it would fix, and — stated with equal clarity — what it would not. The Audit and the Blueprint each get a buy or no-buy, in writing. The do-nothing option is costed too, because sometimes do-nothing is the right call for a quarter, and a briefing that cannot say so is not a briefing.

• A shared vocabulary across engineering, data and finance — the seven stages become the frame the group argues inside instead of past each other
• Your maturity placement, with the reasoning written down
• Written findings — the interview map, the placement and the Core Model sketch — in a one-page summary the sponsor can circulate to stakeholders who were not in the room
• A buy/no-buy on the Audit and the Blueprint the group can defend — including, sometimes, the decision not to buy yet

An enterprise platform group has two agent pilots and a request for seven figures to scale them. The CFO's counter-question — why did the pilot cost triple its estimate? — has no owner. The briefing gives the room one causal story everyone has examined, and the funding conversation restarts from evidence.

A pilot impressed everyone; then an agent answered a customer question from a policy document superseded eight months earlier, and rollout froze. The instinct is to blame the model. The walkthrough usually finds the break earlier in the chain — a sources or validation failure wearing a model costume — which changes what gets funded next.

What is our AI position? is on the next board agenda. A half day produces a placement, a gap read, and a one-pager that answers the question honestly instead of aspirationally.

• It is not a sales presentation. The buy/no-buy is a live output: no-buy and not yet are given in writing when they are true.
• It is not an assessment. No telemetry is reviewed, no retrieval is tested, no content is sampled. The placement is directional; the Core Model Blueprint is where placement becomes measurement.
• It is not a strategy offsite. It is narrow on purpose: one question — where does context break, and is fixing it worth funding — answered well.