The Complete Agentic Office

This is the final exercise of Chapter 39 — and the integration exercise for all of Part 3.

You have built four memory layers. Configured nine skills. Deployed four agents. Learned to produce daily digests, executive dashboards, meeting intelligence, delegation records, and cross-domain context searches. Across twelve lessons, you have constructed every component of the agentic office.

This lesson wires everything together. It reviews what you have built, adds the agent integration layer, runs the integration smoke test, and defines the maintenance cadence that keeps the system accurate and self-sustaining.

"The Digital Chief of Staff is not a single agent. It is the emergent result of four agents working together with a comprehensive work.local.md and all the domain agents from Part 3. When it works, Claude stops being a chatbot you brief every session and starts behaving like a knowledgeable colleague."

This lesson is the test of whether it works.

Capstone Structure

This is a 75-90 minute exercise. Each step builds on the previous. Complete them in order — skipping the review phase means the smoke test will surface gaps you could have fixed in 15 minutes at the start.

Step 1 — Review Your work.local.md (15 minutes)

Before testing anything, verify that work.local.md is complete. Use this checklist:

Section 1: Personal Profile (Layer 1)

• Your role, organisation, and reporting relationships
• Your decision-making style and preferences
• Your communication preferences (format, channel, frequency)
• Your priorities for this quarter (Boulders, in Panaversity's terminology)
• Your working style and context-switching preferences

Section 2: Team Memory (Layer 2)

• At least 5 people entries with name, role, and communication style
• Each entry includes at least: contact preference, current focus, relationship notes
• Dr. Sana Mirza entry updated with onboarding status (joining 17 March)
• Omar Farooq entry includes: lead time preferences, scope clarity requirement
• Ayesha Raza entry includes: new hire context, onboarding window

Section 3: Projects (Layer 3)

• At least 3-5 active projects with codenames
• Each project includes: priority (P1/P2/P3), status, owner, key risks
• AgentFactory: Chapter 39 draft status, Q2 launch target
• Project Nighthawk: current status (escalation in progress), Q3 risk
• BankersAI: Workshop #7 details, content review deadline

Section 4: Organisational Context (Layer 4)

• At least 15 terminology entries (Boulders, Digital FTE, AgentFactory, etc.)
• Meeting rhythm (Executive Weekly, Chapter Review, Banker Workshop)
• Cultural norms and unwritten rules
• Decision-making protocols (CLEAR values, evidence-first, "disagree and commit")
• Communication rules (async-first, Slack/email/WhatsApp routing)

Section 5: Digest Configuration

• digest_time set
• channel configured (where digest is delivered)
• Section order configured (what appears first in digest)

Section 6: Dashboard Configuration

• Domains listed (which domain agents feed the dashboard)
• Refresh frequency configured
• RAG threshold values set (what makes a project go from Green to Amber to Red)

Section 7: Agent Integrations

• chief_of_staff block with schedule and threshold
• memory_keeper block with triggers and staleness thresholds
• meeting_intelligence block with calendar config and deadlines
• work_tracker block with confirmation windows and audit schedule

What to do with gaps: Note the section and what is missing. If a section is largely empty, return to the relevant lesson and complete the exercise before proceeding. An incomplete smoke test is less useful than a 15-minute fix.

Step 2 — Add the Agent Integrations Configuration (10 minutes)

If you completed Step 2 in Lesson 13, verify the configuration is complete. If not, add it now.

The full agent_integrations block connects the four productivity agents to each other and to the domain agents from Chapters 28-38:

agent_integrations:
  chief_of_staff:
    feeds_from:
      - finance_agent: "Budget approvals pending; month-end status"
      - ops_agent: "Compliance dashboard; vendor renewals; change pipeline"
      - sales_agent: "Pipeline; quota; key deal status"
      - hr_agent: "Onboarding status; open positions; compliance training"
      # Add any domain agents you configured in Chapters 28-38
    delivers_to: "Daily digest; executive dashboard"

  memory_keeper:
    triggers:
      - "New person mentioned in conversation"
      - "New project name used"
      - "Meeting completed"
      - "Decision made"
      - "Term used that is not in terminology dictionary"
    updates: "work.local.md — specific section per trigger type"

  meeting_intelligence:
    calendar_integration: "Google Calendar via MCP"
    prep_lead_time: "30 minutes"
    synthesis_deadline: "2 hours"
    always_update:
      - "project status"
      - "decision log"
      - "delegation log"

  work_tracker:
    overdue_threshold_days: 7
    delegation_confirmation_window: 24 # hours
    escalation_path: "digest flag → explicit message → COO-level if 14 days"
    weekly_delegation_audit: "Friday 16:00"

For the chief_of_staff.feeds_from section: add entries for every domain agent you have deployed from Part 3. If you worked through Chapters 28-34, you likely have agents for finance, accounting/CPA, sales/RevOps, and supply chain. Add them here so the Chief of Staff pulls from all of them.

Step 3 — Define Your Trigger Events (10 minutes)

Trigger events define which external occurrences should automatically activate agent actions. These go in a triggers section of work.local.md:

triggers:
  - event: "Meeting ends"
    action: "Meeting Intelligence Agent produces synthesis within 2 hours"

  - event: "New project starts"
    action: "Memory Keeper creates project entry; Work Tracker sets up tracking"

  - event: "Delegation made"
    action: "Work Tracker logs; sets confirmation window; schedules check-in"

  - event: "Regulatory change detected"
    action: "Compliance Monitor alerts; Ops agent updates obligation map"

  - event: "Vendor renewal <90 days"
    action: "Vendor Watchdog alerts; Chief of Staff adds to dashboard"

  - event: "New employee starts"
    action: "HR Onboarding Orchestrator activates; Memory Keeper adds person entry"

Adapt these for your context. If you do not have a sales function, remove the vendor renewal trigger. If you are in a regulated industry, the regulatory change trigger may be the most important one you configure.

Step 4 — Set Your Escalation Thresholds (10 minutes)

For every category of item that can go stale, define the threshold and the escalation action:

escalation_thresholds:
  delegated_task_unconfirmed:
    threshold: 24 # hours
    escalation: "flag in digest → send follow-up message → user decides"

  project_without_status_update:
    threshold: 7 # days
    escalation: "flag in digest → explicit alert → COO-level if 14 days"

  compliance_obligation_overdue:
    threshold: 3 # days before due date
    escalation: "immediate digest flag → explicit alert day-of"

  vendor_renewal_approaching:
    threshold: 90 # days before renewal
    escalation: "add to dashboard → flag in digest 60 days out → urgent 30 days out"

  meeting_action_unclosed:
    threshold: 7 # days
    escalation: "flag in next digest → follow-up in meeting synthesis"

The threshold values above are defaults calibrated for Panaversity's operating pace. Adjust them based on your organisational norms. A fast-moving startup may need half these thresholds; a large enterprise may need double.

Step 5 — The Integration Smoke Test (15 minutes)

Run the following three commands in sequence. For each, grade the output A, B, or C:

• A — Could forward this output directly. Accurate, complete, correctly sourced.
• B — Needs minor editing. Mostly right but missing one or two items.
• C — Needs significant work. Missing key domains, incorrect status, or incomplete sourcing.

Command 1: Executive Dashboard

/agentic-office:progress-tracker

Expected: A full cross-domain RAG dashboard showing status for all active projects and any domain agent feeds configured in agent_integrations. Project Nighthawk should show RED (escalation in progress). AgentFactory should show AMBER (Chapter 39 due Thursday). BankersAI should show GREEN (workshop prep on track).

What to check:

• Are all projects showing?
• Are RAG statuses accurate given the current case study context?
• Are domain agent feeds pulling in? (If finance_agent is configured, is there a budget status line?)
• Is the blocker section populated? (Nighthawk escalation should appear)

Command 2: Daily Digest

/agentic-office:digest

Expected: The full morning digest structure from Lesson 8 — today's priorities, at-risk items, delegation status, meeting prep, and digest notes. Nighthawk should be flagged RED (>7-day threshold). Omar's analytics brief should show confirmation status. The Executive Weekly prep note should appear.

What to check:

• Is the at-risk section showing Nighthawk correctly?
• Is the delegation section showing Omar's brief (midpoint due today)?
• Is the meeting prep section showing the Executive Weekly?
• Is anything present that should not be there?

Command 3: Cross-Domain Intelligence Search

/agentic-office:workplace-search
> What is at risk across all my projects and domains right now?

Expected: A comprehensive cross-domain risk summary drawing from all configured memory layers and domain agent feeds. It should surface:

• Project Nighthawk: facility stall, Q3 timeline risk
• AgentFactory: Chapter 39 deadline Thursday (amber risk)
• Delegation: Ayesha's analysis 5 days without update
• Pending: Banker Workshop content review approaching

What to check:

• Does the answer cover all three active projects?
• Does it pull from both work.local.md and any domain agent feeds?
• Does it distinguish between confirmed risks (Nighthawk) and emerging risks (Ayesha's delegation)?
• Is anything at risk that the search did not surface?

Evaluate Your Results

Record your grades. For each C-grade output:

1. Identify which section of work.local.md is the likely source of the gap
2. Add the missing content
3. Re-run that command
4. Regrade

Most C-grades trace to one of three gaps: Layer 3 (projects not fully populated), agent_integrations.chief_of_staff.feeds_from (domain agents not listed), or digest/dashboard configuration (incomplete threshold or section settings).

Step 6 — Evaluate the Results (15 minutes)

For each output from the smoke test, go deeper than the A/B/C grade:

Question	What You Are Looking For
Is it accurate?	Does the status information match what is actually in work.local.md?
Is it pulling from all sources?	Are all configured domains appearing in the dashboard and digest?
Is anything missing?	What would a knowledgeable colleague include that the output omitted?
Is anything wrong?	Any incorrect status, wrong owner, stale data that should have been updated?
Is it production-ready?	Would you send this output to your COO as-is? If not, what would you change?

The final question — "would you send this to your COO?" — is the real test. Professional-grade outputs from an agentic office should be indistinguishable from outputs a skilled Chief of Staff would produce manually. If your outputs are not at that standard yet, the gap is almost always in the specificity of the underlying work.local.md, not in the agents themselves.

Step 7 — Define the Maintenance Cadence (10 minutes)

The agentic office is only valuable if it stays current. An outdated work.local.md does not just produce less useful outputs — it produces confidently wrong outputs, which are worse than no output at all.

Define your maintenance cadence at four frequencies:

Daily (Automatic)

What the agents handle without intervention:

• Work Tracker: pull open tasks, sort by urgency, flag overdue/unconfirmed
• Chief of Staff: deliver digest at 07:00
• Memory Keeper: trigger on events (new person, new project, meeting, decision)
• Meeting Intelligence: prep brief 30 minutes before significant meetings

Your role: confirm Memory Keeper proposals. Review digest. Act on flags.

Weekly (Manual Review)

What you review every week:

• Monday: Confirm Memory Keeper's weekly maintenance proposals
• Friday: Review Work Tracker's delegation audit — are any patterns worth addressing?
• Friday: Review Meeting Intelligence's weekly audit — any recurring meetings to restructure?
• Update project statuses for any project where status changed this week

Your role: 20-30 minutes every Monday and Friday to confirm and review.

Monthly (Audit)

What you audit once a month:

• Terminology dictionary: any terms added that are redundant, vague, or no longer used?
• People entries: have roles, priorities, or communication preferences changed?
• Project list: are all listed projects still active? Any completed but not closed?
• Decision log: are there decisions that have since been reversed or evolved?

Your role: 30-45 minutes, once a month. Treat it as a standing appointment.

Quarterly (Restructure)

What you restructure every quarter:

• Project priorities: has the P1/P2/P3 assignment still correct given the quarter's progress?
• Organisational structure: has anything changed about your team, reporting, or scope?
• Boulder review: did last quarter's Boulders translate to actual progress? What shifts for next quarter?
• Agent configuration: do the thresholds still match your current operational pace?

Your role: 60-90 minutes, every quarter. This is the equivalent of a quarterly review with a Chief of Staff — the conversation where you recalibrate together.

The Closing Insight

You began Chapter 39 with the Context Problem: every domain agent from Chapters 28-38 was brilliant within its domain and amnesiac across sessions. The finance agent knew nothing about the sales pipeline. The supply chain agent had never heard of Project Nighthawk. Every context switch required a briefing. Every new session started from zero.

What you have built across these fifteen lessons is the solution to that problem. Not by giving agents a longer memory — but by giving them a shared context. One file, four layers, four agents, eleven skills — all pointing at the same work.local.md.

What this means in practice:

When you ask /agentic-office:workplace-search "what is at risk?", it knows about Nighthawk because that project is in Layer 3. It knows the Q3 timeline because the risk is recorded there. It knows Omar's communication preference because that is in Layer 2. It knows that the CLEAR values matter because they are in Layer 4.

The intelligence is not in the agent. The intelligence is in the context the agent has access to. The agent is the synthesiser; work.local.md is the memory.

When the system works correctly — when the Memory Keeper keeps the context current, the Work Tracker ensures nothing falls through the cracks, the Meeting Intelligence captures every decision, and the Chief of Staff synthesises all of it into a coherent operational picture every morning — Claude stops being a chatbot you brief every session and starts behaving like a knowledgeable colleague.

That colleague knows your organisation, your people, your projects, and your priorities. It can tell you what is at risk before you ask. It can draft the escalation message for Nighthawk and know to copy the COO. It can flag that Ayesha is a new hire and that a delayed review sends exactly the wrong signal.

That is the agentic office.

Exercise: The Full Integration

Type: Capstone Integration Time: 75-90 minutes Goal: Complete work.local.md, configure agent integrations, define triggers and thresholds, run and evaluate the smoke test, define maintenance cadence

Work through Steps 1-7 above in sequence. Each step has a specific deliverable:

Step	Time	Deliverable
1. Review work.local.md	15 min	Completed checklist with gaps identified
2. Agent integrations	10 min	Complete agent_integrations YAML block
3. Trigger events	10 min	Triggers section in work.local.md
4. Escalation thresholds	10 min	Escalation thresholds configured
5. Smoke test	15 min	Three command outputs with A/B/C grades
6. Output evaluation	15 min	Gap analysis mapped to work.local.md sections
7. Maintenance cadence	10 min	Four-frequency cadence document

Final deliverable: Complete work.local.md with all seven sections populated. Smoke test results showing at least two A-grade outputs. Written maintenance cadence defining what you will review daily, weekly, monthly, and quarterly.

Try With AI

Try With AI

Use these prompts in Cowork or your preferred AI assistant.

Reproduce: Run the full integration smoke test using case study data.

I want to run a full integration smoke test for an agentic office
configured for Zia Khan at Panaversity.

Context:
- Active projects: AgentFactory (Ch 39 due Thursday, Q2 launch target),
  Project Nighthawk (facility stall 10+ days, Q3 at risk — escalation in progress),
  BankersAI (Workshop #7 content review due in 7 days)
- Delegations: Omar Farooq — analytics ROI brief (due Friday, midpoint check-in today);
  Ayesha Raza — Chapter 38 analysis (due Wednesday, 5 days no update)
- Escalation threshold: 7 days (Nighthawk has breached this)

Run these three commands in sequence and show the expected output for each:

1. /agentic-office:progress-tracker — full dashboard with RAG status
2. /agentic-office:digest — full morning digest for Monday 17 March 2026
3. /agentic-office:workplace-search — "what is at risk across all my projects
   and domains right now?"

For each output, note: what context was required? What is the RAG status for
each project? What would a C-grade version of this output look like
(i.e. what does incomplete work.local.md context produce)?

What you are learning: Running all three commands in sequence shows how the same underlying context (work.local.md) produces three different types of output — structured dashboard, narrative digest, and responsive search. The C-grade comparison is critical: it shows exactly what degrades when memory is incomplete, which is more instructive than seeing a perfect output.

Adapt: Run the same three commands with your own work.local.md and grade each output.

I want to run the integration smoke test with my own professional context.
I will describe my work.local.md content and you will generate the expected
outputs for each command.

My context:
- My role: [Your role and organisation]
- Active projects: [Name, status, priority, key risks for each]
- Delegations: [Who, what, due when, any overdue or unconfirmed]
- Escalation threshold: [N days]
- Domain agents configured: [Which Part 3 domain agents you have deployed]

Generate:
1. /agentic-office:progress-tracker output
2. /agentic-office:digest output (for next Monday morning)
3. /agentic-office:workplace-search — "what is at risk right now?"

After each output, grade it A/B/C:
A = I would forward this directly
B = needs minor editing
C = needs significant work

For each C grade, tell me which section of work.local.md to improve
and what specifically is missing.

What you are learning: The A/B/C grading process forces you to be specific about what "good" looks like for your context. Most people grade their first outputs B or C — and the specific C grades are the most valuable output of the entire capstone exercise, because each one is a direct instruction about where to invest your next 15 minutes of work.local.md improvement.

Apply: Test the system on a Monday morning scenario and evaluate whether it delivers a complete operational picture.

It is Monday morning. I want to test whether a fully configured agentic
office would give me everything I need to start the week without gathering
additional information manually.

Run the full Chief of Staff Monday sequence:
1. Week-ahead brief (06:45)
2. Daily digest (07:00)
3. Executive dashboard (on demand)

Context for this Monday:
- My top priority this week: [State it]
- Biggest risk: [What could go wrong this week]
- Most important meeting: [Name it and what is at stake]
- One delegation I am waiting on: [Who, what, due when]
- One decision I cannot defer: [What is it]

After generating all three outputs, tell me:
- Could I start working immediately after reading them, or would I still
  need to gather information manually?
- What is the single most important piece of context that would improve
  the quality of all three outputs?
- If I could only improve one section of work.local.md today, which
  section would improve the outputs the most?

What you are learning: The Monday morning test is the real-world validation of the agentic office. The question — "could you start working immediately, or do you still need to gather information manually?" — is the test that matters. If the answer is "still need to gather," then the system is not yet at the colleague level. The follow-up questions guide you toward the highest-leverage improvement, which is almost always specificity in one memory layer rather than coverage across all of them.

Flashcards Study Aid

---

Continue to Lesson 15: Summary and Quick Reference →