What is this canvas? · How to run it
This instrument redesigns one business process so that agents execute, people judge and controls audit. It is built for a half-day workshop — about four hours — with every function the process touches in the room and a facilitator driving the screen. Nobody fills it in alone afterwards: the arguments in the room are where the findings come from.
Seven tabs, one per step
- 1 · CANVASGoal: Map the process in nine blocks — trigger, who certifies what, what agents could execute, which judgments it consumes, rules, options, control, staffing, learning.You get: the process on one page, with a delegation backlog and a named auditor.
- 2 · JDEA GRIDGoal: For every task, name who Judges, Directs, Executes and Audits.You get: decisions nobody owns (B3) and tasks ready to hand to agents (J=E), flagged as you type.
- 3 · COVERAGE MAPGoal: Place the process on its lifecycle stage and name a carrier for every judgment it requires.You get: the judgment coverage matrix — empty required cells are staffing gaps.
- 4 · BUG REGISTRYGoal: Scan for the nine known failure modes and note the evidence for any you already see.You get: the failure modes already present, before they cost you.
- 5 · TWISTSGoal: Set the status of each expected role change honestly.You get: which role mutations are emerging and need a plan.
- 6 · SEQUENCEGoal: Give each of the eight implementation steps an owner and a date.You get: an action plan where controls come before any execution moves to agents.
- 7 · REPORTGoal: Read the verdict and the live diagnostics aloud, then export.You get: a process redesign report to circulate as the record of the decisions.
The strip at the bottom of the screen flags problems live — orphaned judgments, audit gaps, delegation candidates, coverage gaps, bugs and sequence progress. Tab 7 turns the whole workshop into the report.
Facilitator view — agenda, timings and notes for each phase →
Step 1 of 7 · Goal: Map the process in nine blocks — trigger, who certifies what, what agents could execute, which judgments it consumes, rules, options, control, staffing, learning. · You get: the process on one page, with a delegation backlog and a named auditor.
The canvas — fill one per process
Work through the nine blocks with every functional seat present. Indigo blocks define human judgment, teal blocks define agent execution, amber blocks define the control layer. Accountability never moves to agents.
Trigger
Calendar, event, or signal? What starts this process today — and what should, once a continuous agent substrate exists?
Accountability map
Which function certifies what? Certification stays human, always.
Execution layer
Assembly, reconciliation, drafting, analysis. Which activities are agent-delegable now / within 12 months?
Judgment layer
Which judgment types does this process consume — taste · trade-off · subtraction · market · risk — and which one person carries each? Carriers are set on the coverage map.
Encoded judgment
Which judgments become agent rules? Every rule needs an owner and a review trigger — unowned rules fossilize (B1).
Option architecture
What option sets should agents pre-generate and pre-validate? Where are the human choice points? Reserve one unscored wildcard slot (prevents B6).
Control plane
Six checks: reconciliation · forecast drift · assumption freshness · override log · option-set hygiene · semantic consistency. Name the auditor.
Staffing shape
Consolidators to redeploy; stewards to create; judgment seats to protect. Stewards before redeployment (prevents B7).
Learning loop
Where do juniors form judgment once execution is absorbed? Override reviews, stuck-session drills, shadow certification (prevents B4).
Step 2 of 7 · Goal: For every task, name who Judges, Directs, Executes and Audits. · You get: decisions nobody owns (B3) and tasks ready to hand to agents (J=E), flagged as you type.
JDEA grid — the task-level actor split
RACI assumed every actor was human. JDEA separates who Judges, who Directs, who Executes (may be non-human), and who Audits.
Rules: every task has exactly one J — one person, never a committee · every task has an A, independent of J and D · an agent needs a D to direct it · J = E means delegation candidate · an empty or shared J is orphaned judgment — bug B3. The grid checks these live.
Seats in this process
People and agents. Seat fields suggest these names; a name not on the list is flagged, and an agent in a human seat (J, D, A, carrier) breaks the rules.| TASK | J — JUDGES | D — DIRECTS | E — EXECUTES | A — AUDITS | DIAGNOSTICS |
|---|
Step 3 of 7 · Goal: Place the process on its lifecycle stage and name a carrier for every judgment it requires. · You get: the judgment coverage matrix — empty required cells are staffing gaps.
Coverage map — lifecycle × judgment type
First place the process/product on its lifecycle stage. Dots show the judgment intensity each stage requires; type the named human carrier in each required cell of your stage. A required cell (●● or more) with no carrier flags GAP. Other stages stay visible for reference — useful when the process is about to transition. Headcount follows judgment gaps — not workload.
| JUDGMENT TYPE | PRE-PMF | GROWTH | MATURE | SUNSET |
|---|
Step 4 of 7 · Goal: Scan for the nine known failure modes and note the evidence for any you already see. · You get: the failure modes already present, before they cost you.
Bug registry — failure modes and their signals
Bugs are failure modes to detect and fix. Each is wired to a control-plane signal so detection is operational, not vibes. Mark a bug detected to log it and open a note field.
Step 5 of 7 · Goal: Set the status of each expected role change honestly. · You get: which role mutations are emerging and need a plan.
Twist registry — role mutations to expect, not fight
Twists are expected mutations to track and support. Treating twists as bugs (resisting role change) while missing real bugs is the classic failure. Set each twist's status as it emerges.
Step 6 of 7 · Goal: Give each of the eight implementation steps an owner and a date. · You get: an action plan where controls come before any execution moves to agents.
Implementation sequence
Order matters: control before delegation, stewards before redeployment.
Step 7 of 7 · Goal: Read the verdict and the live diagnostics aloud, then export. · You get: a process redesign report to circulate as the record of the decisions.
Process redesign report
Generated live from the workspace state — findings, guidelines, and action plan in circulate-able form. Regenerates every time you open this tab. Export produces a self-contained HTML document; print gives PDF.