Complete Community Edition under Apache 2.0. No telemetry.See the source
Agents City

How to run an AI agent committee for a business decision

Compare independent evidence from operations, finance, customer success, and compliance without turning the process into a vote.

Reviewed against Agents City 0.3.0-beta.22 · commit c985198 ↗
01

The real problem

An operations team is deciding whether to automate first-line support. Support logs show volume, finance owns the cost model, compliance owns policy constraints, and operations owns delivery. A single generic prompt hides those evidence boundaries.

02

Target outcome

The operations seat receives isolated positions, synthesises conflicts, records the decision and its conditions, and assigns verification to an independent role.

support-automation-review

Domain
operations
Goal
Decide whether and where to automate first-line support without breaching policy or service targets.
Chair seat
operations-lead
Units and evidence
support-evidencecustomer-success

Ticket themes, escalation patterns, and service impact

cost-modelfin-analytics

Cost assumptions, sensitivity, and expected operational load

policycompliance

Privacy, retention, and approval constraints

deliveryprogram-manager

Rollout sequence, dependencies, and rollback plan

04

Before you start

  1. 01

    Version-controlled folders or repositories containing the evidence each role should inspect.

  2. 02

    A named human decision owner and a question with explicit authority and constraints.

  3. 03

    A verifier that did not author the chosen execution plan.

05

How to do it

  1. Step 01

    Create an operations city

    Choose the operations domain and operations-lead seat role. Set a goal that names the decision and the non-negotiable service and policy constraints.

    TERMINAL
    agents-city cities create support-automation-reviewagents-city seat --city support-automation-review --domain operations --role operations-lead
  2. Step 02

    Attach evidence, not generic personas

    Select the support, finance, policy, and delivery repositories. Assign roles based on the evidence each unit owns; cross-domain roles remain member agents.

    TERMINAL
    agents-city seat --city support-automation-review --reposagents-city seat --city support-automation-review --agent-roles
  3. Step 03

    Ask for isolated positions

    Write a committee brief that asks each participant for evidence, risks, unknowns, and a recommendation within its scope. Do not ask agents to vote on the final decision.

    TERMINAL
    agents-city committee schema openagents-city committee open --input support-automation.json
  4. Step 04

    Record conditions and independent verification

    The chair synthesises the evidence and records one decision. Include the executor, dissent, rollout conditions, and the evidence that would reopen the act.

    TERMINAL
    agents-city committee show <deliberation-id>
06

What you should get

  • The decision separates source evidence from synthesis and final authority.
  • Material disagreement remains visible instead of disappearing into one generated answer.
  • The act cannot close successfully until an independent verification passes.
07

What this does not solve

  • Agents City is not a substitute for qualified legal, financial, medical, or compliance review.
  • Bad or incomplete evidence produces a traceable process, not a magically correct decision.
  • The human owner remains responsible for the business decision and its consequences.