The Agent Department Org-Design Worksheet
The triad (PM, Designer, Engineer) is dead. Pods are dead. The new unit isn't humans-with-agents-bolted-on; it's a human department coordinating with an agent department on exception.
The template
The Agent Department Org-Design Worksheet
A template for designing an agent department alongside a human department.
By Falk Gottlob — falkster.com
The premise
The triad (PM, Designer, Engineer) is dead. Pods are dead. The new unit isn't humans-with-agents-bolted-on; it's a human department coordinating with an agent department on exception.
This worksheet maps the agents you have to the human departments they coordinate with, and surfaces the gaps where coordination is missing.
Step 1: List the human departments
| Human Department | Lead | Headcount | Mandate |
|---|---|---|---|
| Product | |||
| Engineering | |||
| Design | |||
| Sales | |||
| Customer Success | |||
| Marketing | |||
| Finance |
Step 2: Map the agent departments
For each agent department, list: agents that compose it, the workload it handles, the human department it coordinates with, the escalation path.
Agent Department: Discovery
- Agents: continuous-listening, interview-synthesis, customer-segmentation, journey-mapping, NPS/CSAT analysis
- Workload: continuous customer signal ingestion, interview transcript synthesis, qualitative pattern detection
- Coordinates with: Product (for prioritization), CCO (for escalations)
- Escalation: when signals diverge across sources, or a customer-named risk surfaces, escalate to PM lead
Agent Department: Quality
- Agents: kpi-watchdog, agent-quality-scorer, eval-runner, drift-detector, dispute-classifier
- Workload: continuous monitoring of product KPIs and agent quality, eval pipelines, dispute triage
- Coordinates with: Engineering (for fixes), CS (for customer-facing disputes), CPO (for cross-cutting trends)
- Escalation: when eval pass rate drops below threshold for 7 days, or agent quality score drops below 3.5/5
Agent Department: GTM
- Agents: competitive-intel, market-intel, win-loss-analysis, gtm-monitoring
- Workload: competitive surface scanning, win/loss synthesis, GTM dashboard maintenance
- Coordinates with: Marketing, CRO, Product
- Escalation: when competitive moves materially affect positioning, or GTM metrics drift outside band
Agent Department: Operations
- Agents: product-ops, weekly-ops-digest, retrospective-synthesis, stakeholder-communication, executive-report
- Workload: ops cadence (weekly digests, retros, exec reports, stakeholder updates)
- Coordinates with: Every human department
- Escalation: when a ritual breaks down (a weekly digest doesn't happen, a retro reveals systemic issues)
Add your own
The four above cover most of the use cases I see. Your company may have more (a Talent agent department, a Compliance agent department, a Pricing agent department). The frame is the same.
Step 3: The coordination gaps
For each agent department, ask:
- Is there a clear human owner who reads the agent's output every day?
- Is there a clear escalation path when the agent flags an issue?
- Is there a dispute mechanism when the agent and the human disagree?
- Is there a quarterly review of whether the agent department is still scoped right?
If any answer is "no," the agent department isn't really running. It's a set of agents producing outputs nobody owns.
Step 4: The bridge function
Some agents need cross-department coordination. Define the bridge function explicitly.
| Bridge | Owner | Cadence | What it surfaces |
|---|---|---|---|
| Quality + GTM | Senior PM | Weekly | When quality issues affect win rate |
| Discovery + GTM | Product Marketing | Bi-weekly | When customer signals affect positioning |
| Operations + Quality | VP Eng | Weekly | When ops rituals reveal recurring quality issues |
What this worksheet doesn't do
- It doesn't tell you which agents to build first. Build the ones that solve the most repetitive human work in your specific company.
- It doesn't replace human ownership. Every agent department needs a human owner who reads its output daily.
- It doesn't work if you treat it as a one-time exercise. Rerun every six months as the agent fleet changes.
This worksheet pairs with The PM AI Agent Fleet and PM-as-Editor.