You finish with a mission brief, permission matrix, approved plan and the record of one small, reversible run.
Agentic work: delegate with limits and evidence
The official PDF remains central. This guide turns its fifteen slides into operational decisions and does not replace the document.
- 15 · Moments
- SHA-256
710cff50558a…
Official masterclass PDF
The official PDF is not tagged. Use the accessible HTML guide to follow its content.Delegate a process without giving up your judgment.
Giving an AI a vague goal can accelerate the wrong outcome. Useful autonomy begins by stating what it may decide, what it must escalate and how success is checked.
Choose a case, design the contract, limit tools, approve the plan, observe execution and decide whether to keep, adjust or stop.
Keep the brief, approved plan, exceptions, sources used, result and final decision.
Moments
Follow the content moment by moment. This guide accompanies the PDF; it does not replace its content or design.
- Core route
- 90 min
- With extended practice
- 120 min
Document note
The official PDF remains central. This guide turns its fifteen slides into operational decisions and does not replace the document. The official PDF is image-based and has no reading-order tags. Use this HTML guide for a text version of the sequence, ideas and practices.
Moments01 / 15
01 · Understand
Distinguish assistance, delegation and orchestration without hiding accountability or coordination.
02 · Design
Turn the case into a Mission Brief, with sources, permissions, human decision points (gates) and a Definition of Done.
03 · Pilot
Approve, observe and review a reversible run before deciding continuity. In this pilot, no external effect is authorized.
From creator to orchestrator
Delegation starts by turning known work into a bounded objective, not by asking the agent to “handle everything.” Your case keeps a verifiable personal boundary.
Go deeperUnderstand
Key points
- Name the intended result.
- Make exclusions part of the contract.
- Application
- Define the result and one exclusion.
- Check your understanding
- Can you state what completes the mission and what does not belong to it?
- Short practice
- In the synthetic personal example, write one sentence with objective, boundary and done condition.
- Keep this idea
- Bound the mission before choosing what to delegate.
- Connect to the next resource
- Take the case to the suitability matrix.
From productivity to leverage
Doing a task faster is different from designing a repeatable, reviewable cycle.
Go deeperUnderstand
Key points
- Speed describes one run.
- A system includes inputs, decisions and evidence.
- Application
- Draw the cycle that should repeat.
- Check your understanding
- Could the next run follow the same contract?
- Short practice
- In the synthetic personal example, write input, decision, output and evidence.
- Keep this idea
- Leverage requires an observable system, not a promise.
- Connect to the next resource
- Use the cycle to locate its bottleneck.
Find your bottleneck
The bottleneck may be a necessary human decision; removing it does not always improve the flow.
Go deeperUnderstand
Key points
- Separate avoidable waiting from essential judgment.
- Keep approval where it protects the result.
- Application
- Mark where judgment is required.
- Check your understanding
- Which decision should the agent never assume?
- Short practice
- In the synthetic personal example, underline one point that needs human review.
- Keep this idea
- Optimization does not mean erasing accountability.
- Connect to the next resource
- Compare that point with co-creation and delegation.
Co-creation is not delegation
Co-creation keeps step-by-step interaction; delegation assigns a bounded mission with agreed review. You choose when to converse and when to review.
Go deeperUnderstand
Key points
- Cadence and contract change.
- Accountability does not move to the system.
- Application
- Classify the case as assistance or mission.
- Check your understanding
- Do you need dialogue at every step or review of a defined result?
- Short practice
- In the synthetic personal example, rewrite one task as an objective and keep one gate.
- Keep this idea
- Delegation changes the workflow, not final accountability.
- Connect to the next resource
- Make the required coordination explicit.
Scale without hiding coordination
Agentic work still requires supervision, dependencies and exception handling. Separate your boundaries from external dependencies.
Go deeperUnderstand
Key points
- Coordination still has a cost.
- Every dependency needs an owner or stop condition.
- Application
- Name one dependency and one exception.
- Check your understanding
- Who responds when an input is missing or a conflict appears?
- Short practice
- In the synthetic personal example, add an exception with a stop-or-escalate response.
- Keep this idea
- Do not count shifted supervision as eliminated work.
- Connect to the next resource
- Move from the idea to the capabilities you will authorize.
Capabilities behind agentic delegation
Context, memory, tools and actions are separate permissions, not one bundle. Authorize only your own data and tools.
Go deeperDesign
Key points
- Authorize only what the case needs.
- Undeclared access remains prohibited.
- Application
- Separate reading, proposing and acting.
- Check your understanding
- What is the minimum permission needed for a reviewable draft?
- Short practice
- In the synthetic personal example, create three columns: may read, may propose, may not do.
- Keep this idea
- Useful capability is specific and revocable.
- Connect to the next resource
- Connect permissions to the work cycle.
From objective to work cycle
A governed cycle defines input, decision, action, evidence and stop before execution.
Go deeperDesign
Key points
- Every step produces a reviewable signal.
- The stop condition prevents momentum from becoming authority.
- Application
- Order the five parts of the cycle.
- Check your understanding
- What evidence allows the next step?
- Short practice
- In the synthetic personal example, write one advance condition and one stop condition.
- Keep this idea
- Without visible states, execution cannot be reviewed.
- Connect to the next resource
- Place human controls inside the cycle.
Architecture with human control
Human gates sit before sensitive actions and when results require acceptance. Your approval retains real stop authority.
Go deeperDesign
Key points
- Plan approval is not approval of every effect.
- A reviewer needs criteria, not only an alert.
- Application
- Choose one gate before and one after.
- Check your understanding
- What must a person see to approve or reject?
- Short practice
- In the synthetic personal example, write a verdict: approve, correct or stop.
- Keep this idea
- Human control works when it has information and power to stop.
- Connect to the next resource
- Separate deliverable quality from system quality.
Craftsperson and orchestrator
Producing a strong deliverable and designing a sound system are related but distinct jobs.
Go deeperDesign
Key points
- The craft role protects the output.
- The orchestration role protects rules, evidence and continuity.
- Application
- Assign one review to each role.
- Check your understanding
- What belongs to output review and what belongs to process review?
- Short practice
- In the synthetic personal example, write one deliverable criterion and one system criterion.
- Keep this idea
- Quality requires review of both result and mechanism.
- Connect to the next resource
- Turn both criteria into a mission brief.
The mission brief as contract
The brief makes the mission testable through role, objective, context, limits and done. The Mission Brief is the bounded contract for scope and acceptance.
Go deeperDesign
Key points
- An objective without boundaries invites improvisation.
- Done criteria allow acceptance or rejection.
- Application
- Complete the five brief elements.
- Check your understanding
- Could someone outside the case tell what to approve?
- Short practice
- In the synthetic personal example, replace one vague verb with observable evidence.
- Keep this idea
- The mission brief is the pilot's operating contract.
- Connect to the next resource
- Complete it with sufficient sources.
Role, objective and context
A mission needs identified sources and an explicit statement of what is missing. Use only personal or authorized sources.
Go deeperDesign
Key points
- Broader access does not fix a missing source.
- A material gap must stop or escalate.
- Application
- List sources and gaps separately.
- Check your understanding
- Which missing fact would change the plan or verdict?
- Short practice
- In the synthetic personal example, mark each source available, limited or absent.
- Keep this idea
- Honesty about gaps protects the decision.
- Connect to the next resource
- Use the inventory to restrict tools and instructions.
Instructions, tools, limits and done
Instructions must state allowed steps, tools, prohibitions, escalation and done.
Go deeperDesign
Key points
- A tool never authorizes every possible use.
- Every prohibition needs a safe response.
- Application
- Turn one prohibition into a stop condition.
- Check your understanding
- Does the agent know what to do when it cannot proceed?
- Short practice
- In the synthetic personal example, add “stop and request review” to one exception.
- Keep this idea
- Useful limits state both what not to do and how to respond.
- Connect to the next resource
- Send the contract to the plan gate.
Approve the plan before execution
Before execution, review sequence, sources, risk, gates and reversibility of the proposed plan. You compare the plan with your boundaries before approval.
Go deeperPilot
Key points
- The plan is a proposal subject to approval.
- A missing source or permission blocks the affected step.
- Application
- Audit the plan before authorizing it.
- Check your understanding
- Is every action backed by a source and permission?
- Short practice
- In the synthetic personal example, mark one step to approve and one to correct.
- Keep this idea
- Approving the objective does not replace approving the plan.
- Connect to the next resource
- Reduce the plan to a small, recoverable mission.
Design a bounded mission
The first pilot should be small, frequent, observable and recoverable if it fails. In this pilot, you do not send, publish, pay or change external systems.
Go deeperPilot
Key points
- Start without publishing, paying or sending.
- Test the path and evidence first.
- Application
- Choose a controlled case and restoration path.
- Check your understanding
- Can you return to the prior state without relying on the agent?
- Short practice
- In the synthetic personal example, define a dry run and how you would restore state.
- Keep this idea
- The first mission demonstrates control, not scale.
- Connect to the next resource
- Record the pilot to decide continuity.
Turn practice into a governed asset
The pilot earns value through an informed keep, adjust or retire decision. Your verdict chooses keep, adjust or retire.
Go deeperPilot
Key points
- Compare evidence with done.
- Version changes and retain exceptions.
- Application
- Close with a verdict and next gate.
- Check your understanding
- What evidence supports continuation, and what requires stopping?
- Short practice
- In the synthetic personal example, write a decision with reason, adjustment and review date.
- Keep this idea
- Continuity is earned through evidence, not enthusiasm.
- Connect to the next resource
- Transfer the verdict to the playbook and next mission.
Workbook
Continue with the next resource in the journey.
Masterclass → Workbook