Synthetic sample pre-read
What participants receive before the Executive Workshop: the three levers in brief, one worked example, and a ten-minute questionnaire whose answers shape the scenarios. Everything here is illustrative and synthetic — the real pre-read is tailored to your initiatives from the questionnaire.
The three levers, in one page
Rewards and objectives
What the workflow is optimising, and how that objective reads under pressure. Before the session, think of one metric in your organisation that is pursued more precisely than it is understood.
Explicit prohibitions and constraints
What the workflow must never do, expressed where it can actually be enforced. A control that is not enforced is documentation.
Earned authority: permissions and escalation
How much authority the workflow has earned, and what evidence would justify more (or less). Authority grows with evidence, not enthusiasm.
One worked example: what a checkable constraint looks like
In a prompt (not a control)
"Never contact customers directly; keep to the approved pricing model."
As a checkable constraint
The workflow's service role has no send permission; the pricing tool only returns values validated against the approved model, and the release step refuses to assemble a proposal without that validation.
The session works this contrast on your own workflows (synthetic example shown).
The ten-minute questionnaire (sample questions)
- Which AI initiative in your organisation is closest to a business-critical workflow?
- Who could stop it tomorrow if it misbehaved — by name?
- What evidence convinced the sponsor it was working: a demo, or a measurement?
- What is the workflow not allowed to do? Where is that written, and what enforces it?
- What would you need to see before granting it one more step of authority?
Vocabulary we will use
- Boundary
- What the workflow may do, enforced outside the model.
- Blast radius
- What a single tool call can reach or change if the agent is manipulated.
- Acceptance criteria
- What "good enough to ship" means, in testable terms.
- Exception rate
- Share of tasks handed to a human because confidence or policy was not met.
- Earned authority
- Permissions granted on evidence, reviewed like any access.
- Escalation
- A named human receiving a decision, not an error log nobody reads.
The session starts where the pre-read ends.
Three hours, your scenarios, your shortlist — facilitated by a principal.