Skip to content

Agent operating model

Give agents the task.
Keep responsibility human.

Give each agent a clear task, the context it needs, and an expected output. Every execution returns to a person for review with its model route, actions, and result attached.

See the delegation flow

01TaskOne bounded piece of work

02ContextOnly what the task needs

03ReviewA person accepts the result

Execution lifecycle

Assign the work before the agent starts.

The task carries the outcome, context, limits, and acceptance conditions. The agent executes inside that contract and returns the result to the same delivery path.

  1. 01

    Define the role contract

    Give each agent a purpose, capabilities, constraints, expected outputs, tools, provider route, and memory boundary.

  2. 02

    Attach work to a task

    An agent joins a Goal through assigned work. The task carries the outcome, dependencies, owner, context, and acceptance conditions.

  3. 03

    Provide scoped context

    Attach the files, notes, links, and approved Knowledge needed for this task instead of exposing an entire workspace by default.

  4. 04

    Review the output

    Agent output remains connected to its execution and source task. People can inspect, revise, reject, or approve the result.

Clear responsibility

Agents execute. People decide what counts.

Agents can

Run assigned tasks, use approved tools and context, draft outputs, and return questions when the work cannot continue safely.

People own

Consequential choices, reviews, approvals, accepted deliverables, and the decision to close the Goal.

Review the execution boundary

Test a bounded agent task

See the agent contract operate on real delivery work.

Bring one task, its context, provider route, expected output, and human review point.