Manager Pathway
Module B1 · 45 min

Practical Use Cases — ZHD & PROBE

Two end-to-end agentic systems that managers will actually meet. The Zero Human Department for market intelligence shows what governance looks like when no daily human staff sit inside the function. PROBE shows what a structured multi-agent pipeline looks like when every phase transition needs operator authorisation.

Free with account
Objectives
  • Govern a Zero Human Department without operating it day-to-day.
  • Read a PROBE pipeline and identify every phase-boundary intervention point.
  • Distinguish Monitor / Verify / Override / Escalate as discrete operator modes.
  • Map each use case to the HIMM dimensions it stresses.
Use Case I · Zero Human Department

Market Intelligence without a market intelligence team

A multi-agent system that scouts signals, synthesises intelligence, evaluates its own outputs, maintains memory across cycles, and delivers a fresh briefing — autonomously. One human operator governs from outside.

Architecture map
01Human Operator · Governance layerverify
Role

You set direction, review outputs, handle exceptions, answer for outcomes.

Operator obligation

Verify weekly intelligence reports before distribution; monitor evaluation scores; escalate when quality drops below threshold.

02Orchestrator · Directorverify
Role

Decomposes operator objectives into scout tasks, dispatches in parallel, consolidates results.

Operator obligation

Review and approve the operator brief at the start of each cycle. Investigate repeated recycling — it signals drift in synthesis or evaluation.

03Scout agents (Web · News · Regulatory · Social · Competitor)monitor
Role

Parallel specialist scouts that gather signal by domain.

Operator obligation

Monitor coverage weekly. Regulatory high-impact findings reach you directly and require verification before synthesis.

04Synthesiser + Self-evaluator + Publisheroverride
Role

Synthesises briefs, scores its own output, publishes only after the evaluator clears it.

Operator obligation

Verify the final brief after evaluation — your primary governance touchpoint. Halt publication on threshold failure.

05Memory (Episodic · Semantic · Working)monitor
Role

Shared infrastructure: what happened, what things mean, this cycle's context.

Operator obligation

Audit semantic memory quarterly — outdated definitions silently corrupt every downstream brief.

Operator obligations
ModeTriggerAction
MonitorSelf-evaluation scores, weeklyReview aggregate quality. If average drops below your threshold (e.g. 7/10), pause and investigate.
VerifyWeekly intelligence report, before distributionRead the synthesised report. Apply domain judgment: is anything missing, biased, factually suspect?
VerifyRegulatory signal flagged high-priorityVerify directly before the synthesiser processes it. Significance is not delegable.
OverrideOutput fails quality threshold or contains material errorHalt publication. Return the brief with a specific correction instruction. Document the error type.
OverrideSelf-evaluator consistently approves low-quality outputRecalibrate scoring criteria. The evaluator is drifting; this is your call, not the system's.
HIMM dimensions stressed
D1D2D3
Linked assets