Persistent Agents — Onboarding Orchestrator and Policy Maintenance
It is three days before Ayesha Raza's start date. On paper, the pre-boarding checklist is progressing. The offer letter was signed three weeks ago. The IT system provisioning request was submitted. But nobody checked whether the laptop had actually been ordered. It had not — the IT provisioning request had been submitted to the wrong shared inbox, silently misfiled, and forgotten. Nobody noticed, because nobody was looking.
Ayesha's Day 1 has her starting at 9am, working through an IT setup session, and accessing the data warehouse by lunchtime. Instead, she spends Day 1 on a laptop borrowed from a colleague, with no data access, no system credentials, and an IT team scrambling to catch up on a provisioning request that should have been actioned two weeks ago. By Week 1, she is behind. By the 30-day check-in, the initial impression is more complicated than it should be.
The Onboarding Orchestrator would have caught this. At T-3 days, the agent checks the pre-boarding checklist against completion status and identifies any critical items still incomplete. "IT provisioning not confirmed — laptop not ordered" would have generated an immediate escalation to the HR team, three days before Ayesha's start date. Three days is enough time to fix it. One hour on Day 1 is not.
This lesson is about the two persistent agents that run continuously in the background of an AI-native HR operation — catching the things humans miss because humans are busy with the 60% of administrative work that these same agents help reduce.
Two Agent Architectures
The four agents in the hr-operations plugin follow two distinct architectures:
| Architecture | Trigger | Examples |
|---|---|---|
| Event-triggered | Activated by a specific HRIS event | onboarding-orchestrator (new hire record), offboarding-knowledge-agent (resignation record) |
| Scheduled | Activated on a fixed cadence | policy-maintenance-agent (monthly, 1st Monday) |
| Hybrid | Scheduled cadence + event override | policy-maintenance-agent (monthly + statutory rate change event) |
The knowledge-base-agent is a third pattern — always-on, responding to individual employee queries — covered in Lesson 4.
Understanding which architecture fits a use case determines how the agent is configured. Event-triggered agents need an HRIS integration that fires when a record is created or changed. Scheduled agents need a cron schedule and a defined monthly workflow. Hybrid agents need both.
The Onboarding Orchestrator
Architecture: Event-triggered (HRIS new hire record) Sensitivity: CONFIDENTIAL What it catches: Everything humans forget when they are managing five other new starters
The T-14 to Day 90 Timeline
When a new hire record is created in the HRIS, the orchestrator activates. It then runs a structured workflow at eight milestones:
| Milestone | Agent action | Who receives it |
|---|---|---|
| T-14 days | Generate pre-boarding checklist; assign tasks to HR, IT, and Manager | HR team, IT team, Manager |
| T-7 days | Send Day 1 schedule to new starter; chase any incomplete pre-boarding items | New starter, HR, Manager |
| T-3 days | Check all critical pre-boarding items; escalate any incomplete items immediately | HR team (urgent alert) |
| Day 1 | Confirm IT setup complete; send personalised welcome message from HR | New starter, Manager |
| Day 10 | Check mandatory compliance training completion; chase incomplete | Manager, new starter |
| Day 30 | Schedule 30-day check-in; send check-in agenda to manager and new starter | Manager, new starter |
| Day 60 | Schedule 60-day check-in; gather feedback from new starter (satisfaction survey) | Manager, new starter, HR |
| Day 90 | Schedule 90-day review; generate 30-60-90 completion summary | Manager, new starter, HR Business Partner |
Three Alert Triggers
The orchestrator is not just a reminder system. It has three alert conditions that escalate to HR immediately — not at the next scheduled milestone:
Alert 1 — T-3 incomplete critical pre-boarding:
Any of these items still incomplete at T-3 days triggers an immediate HR alert:
- Laptop not ordered or not confirmed delivered
- System access provisioning not completed
- Day 1 schedule not sent to new starter
- Manager unavailable on Day 1 (no coverage arranged)
Why: these items cannot be fixed in the final three days if caught on Day 1.
Alert 2 — Mandatory training incomplete at Day 10:
If the new starter has not completed mandatory compliance training (GDPR, information security, code of conduct) by Day 10, the agent escalates to the manager with a specific deadline. Mandatory training is not optional and not flexible — it is a compliance requirement.
Why: most training non-completion in the first 30 days is not intentional — it is an admin failure. The escalation turns "the training email got buried" into a tracked, owned item.
Alert 3 — Satisfaction survey score at or below threshold:
The Day 60 satisfaction survey produces a score. If the new starter rates their onboarding experience at or below the configured threshold (typically 2 out of 5), the orchestrator escalates to the HR Business Partner for a direct check-in conversation.
Why: a 2/5 score at Day 60 is a retention signal, not just a satisfaction signal. The HRBP check-in gives the new starter a confidential channel to surface concerns before they make a decision to leave.
Ayesha Raza — Tracing the Full Workflow
Ayesha Raza joins the EdTech company as Senior Data Analyst, start date 10 March 2026. Here is what the orchestrator does:
ONBOARDING ORCHESTRATOR: Ayesha Raza — Senior Data Analyst
Start date: 10 March 2026 | Manager: Omar Farooq
================================================================
T-14 (24 February):
→ Pre-boarding checklist generated and assigned:
HR: Offer letter filed ✓, right-to-work verified ✓,
benefits enrolment sent ✓, Day 1 schedule drafted ☐
IT: Laptop order initiated ☐, system access provisioning ☐,
email set up ☐
Manager: Team introduction email ☐, buddy assigned ☐,
first-week schedule blocked ☐
T-7 (3 March):
→ Day 1 schedule sent to Ayesha
→ Pre-boarding status check:
⚠️ IT: Laptop order confirmed. System provisioning not completed.
⚠️ Manager: Team introduction email not sent yet.
→ Chase sent to IT and Omar
T-3 (7 March):
→ Pre-boarding status check:
✅ Laptop confirmed shipped, arrival Day 1 AM
✅ System access provisioning complete
⚠️ ALERT: Manager Day 1 availability — Omar in external meeting
8:30–10am. Day 1 welcome handover arranged: team lead covers.
→ HR alert sent with Omar's morning conflict flagged
Day 1 (10 March):
→ IT setup confirmed complete ✓
→ Welcome message sent to Ayesha from HR
→ Buddy confirmed: assigned data engineer
Day 10 (20 March):
→ Mandatory training check:
✅ GDPR training: complete (14 March)
✅ Information security: complete (16 March)
⚠️ Code of conduct sign-off: not yet completed
→ Reminder sent to Ayesha with link; deadline: 25 March
Day 30 (9 April):
→ 30-day check-in scheduled: 14 April, Omar + Ayesha
→ Check-in agenda sent to both
→ Onboarding status: all mandatory training complete ✓
Day 60 (9 May):
→ 60-day check-in scheduled: 13 May
→ Satisfaction survey sent to Ayesha
→ Survey result: 4.2/5 (above threshold — no alert)
→ HR copy of survey result filed
Day 90 (8 June):
→ 90-day review scheduled: 12 June, Omar + Ayesha + HR Business Partner
→ 30-60-90 completion summary generated:
All mandatory training: complete
Check-ins completed: Day 30 ✓, Day 60 ✓
Survey result: 4.2/5
Satisfaction trend: stable
→ Summary sent to Omar and HR Business Partner as pre-read
================================================================
The orchestrator does not judge whether Ayesha is performing well. That is Omar's job. It ensures that the administrative infrastructure of onboarding — the checklist completion, the training, the check-ins, the surveys — actually happens, with someone accountable for every item.
The Policy Maintenance Agent
Architecture: Hybrid (monthly schedule + event-triggered on statutory rate changes) Sensitivity: ROUTINE What it catches: Policy drift, out-of-date rates, broken links, FAQ gaps
Five Monthly Checks
On the first Monday of each month, the policy maintenance agent runs five checks across the organisation's HR documentation:
| Check | What it looks for | Example finding |
|---|---|---|
| 1. Policy version currency | Are all documents referencing the current policy version? | "Employee handbook still references the 2024 holiday entitlement; 2025 version updated the accrual rate" |
| 2. Statutory rate monitoring | Are all rates current for the configured jurisdiction? | "Statutory Sick Pay rate in FAQ: £109.40/week. Current rate: £116.75/week [VERIFY] — update required" |
| 3. Document consistency | Is the same policy described the same way in different documents? | "Parental leave section: handbook says 26 weeks; intranet FAQ says 24 weeks — inconsistency flagged" |
| 4. Link validity | Are all policy links in employee-facing documents still active? | "3 broken links in the benefits guide — pages moved after HRIS migration" |
| 5. FAQ gap analysis | Are there recurring question categories in the knowledge-base-agent log that suggest a FAQ entry is missing? | "14 queries this month about jury duty leave — no FAQ entry exists" |
Event-Triggered: Statutory Rate Changes
In addition to the monthly cycle, the policy maintenance agent monitors for statutory rate changes in the configured jurisdiction. When a change is detected, it triggers an immediate (not monthly) alert:
UK jurisdiction monitoring:
- National Living Wage: changes April each year
- Statutory Sick Pay: changes April each year
- Statutory Maternity/Paternity Pay: changes April each year
- The agent monitors the relevant government sources and flags changes before they take effect, giving HR time to update policies before the new rate applies
Pakistan jurisdiction monitoring:
- Provincial minimum wages: vary by province; change periodically
- The agent monitors the applicable provincial authority and flags changes as they are announced
Event-triggered workflow:
Rate change detected: National Living Wage — effective 6 April 2026
↓
Agent searches all employee-facing documents for the current rate
↓
Finds: Employee handbook (p.12), FAQ entry #7, offer letter template
↓
Alert to HR: "NLW rate change detected. These 3 documents contain
the current rate and require updating before 6 April:
[list with location and specific text to change]"
↓
HR confirms updates completed → agent verifies and files confirmation
Sample Monthly Maintenance Report
HR POLICY MAINTENANCE REPORT
Month: March 2026 | Generated: 1 March 2026 (first Monday)
Jurisdiction: UK | Configuration: hr.local.md loaded
POLICY VERSION CURRENCY
✅ Employee handbook: current version (v2026.1)
⚠️ Benefits guide: references 2024 pension contribution rates —
PAYE rates updated January 2026; needs refresh
✅ Remote working policy: current
STATUTORY RATE MONITORING
⚠️ IMPORTANT: National Living Wage changes 6 April 2026.
Current rate in documents: £11.44/hour
New rate: [VERIFY — check gov.uk before publication]
Documents to update: offer letter template, FAQ entry #3,
employee handbook section 8.2
✅ Statutory Sick Pay: correct rate in all documents
DOCUMENT CONSISTENCY
⚠️ INCONSISTENCY: Parental leave entitlement
Employee handbook (Section 5): "26 weeks at full pay + 13 at statutory"
HR intranet FAQ: "Up to 26 weeks full pay"
→ "Up to" is ambiguous and potentially incorrect; align with handbook
LINK VALIDITY
❌ 3 broken links detected:
Benefits guide, page 4: pension enrolment link (broken since Feb migration)
Benefits guide, page 8: private health portal link (URL changed)
Remote working policy, section 3: IT equipment request link (404)
FAQ GAP ANALYSIS (from knowledge-base-agent log)
New question categories requiring FAQ entries:
— Jury duty leave (14 queries this month; no FAQ exists)
— Work from abroad requests (9 queries; current FAQ is ambiguous)
RECOMMENDED ACTIONS (prioritised)
1. [URGENT] Verify and update NLW rate in 3 documents before 6 April
2. Fix 3 broken links in benefits guide and remote working policy
3. Add jury duty leave FAQ entry
4. Clarify parental leave entitlement wording (align with handbook)
5. Refresh benefits guide pension contribution rates
6. Clarify remote work abroad FAQ — add policy position clearly
NEXT REPORT: 6 April 2026 (first Monday of April)
The specific policy issues, rates, and documents will depend on your organisation's jurisdiction, documentation state, and the knowledge base agent query log. The structure of the report — five check categories, prioritised actions — is consistent regardless of what the agent finds.
Exercise: Configure Both Agents and Run a Simulated Monthly Audit
Type: Configuration and Applied Testing
Time: 40 minutes
Plugin: onboarding-orchestrator + policy-maintenance-agent (deploy from hr-operations plugin)
Goal: Configure both persistent agents and run a simulated monthly policy audit to experience the maintenance workflow
Step 1 — Configure the Onboarding Orchestrator
In Cowork, deploy the onboarding-orchestrator from the hr-operations plugin. Configure it with:
New hire: Ayesha Raza (from Chapter scenario) or a new hire of your choice
Start date: [Date — any upcoming Monday works]
Manager: Omar Farooq
Jurisdiction: Pakistan
Mandatory training: GDPR, Information Security, Code of Conduct
Satisfaction survey threshold: 2.0 out of 5.0
HR alert contact: [Your name or fictional HR contact]
Then trace the full workflow: for each of the eight milestones (T-14 through Day 90), write one sentence describing what the agent sends, to whom.
Extend: Simulate the T-3 scenario — mark the laptop as "not confirmed" in the pre-boarding checklist. Verify the agent generates an alert. Who receives it? What does it say?
Step 2 — Configure the Policy Maintenance Agent
Deploy the policy-maintenance-agent from the hr-operations plugin. Configure it for:
Jurisdiction: UK (or Pakistan — your choice)
Policy documents to monitor: [List 3 policies — can be fictional]
1. [Policy name + location]
2. [Policy name + location]
3. [Policy name + location]
HR contact for alerts: [Your name or fictional HR contact]
Monthly schedule: First Monday of each month
Statutory rate monitoring: Enabled
Step 3 — Run a Simulated Monthly Audit
Ask the policy maintenance agent to simulate a monthly audit on this fictional policy set:
Please run a simulated monthly policy audit for the following
HR documentation set. For each of the five check categories, report
your findings and identify any issues requiring action.
Documents:
1. Employee Handbook (fictional company, UK jurisdiction)
— Sick pay: 10 days full pay, then SSP
— Annual leave: 25 days per year
— Parental leave: 26 weeks full pay (primary carer)
2. HR FAQ document (last updated January 2025)
— FAQ #1: "How do I report sick?" (references old process)
— FAQ #4: "What is the SSP rate?" [Note: uses £109.40 — old rate]
— No FAQ entry for jury duty leave
3. Benefits guide
— Two links to pension provider portal (one may be broken)
— References "2024 pension auto-enrolment thresholds"
Run the five monthly checks and produce a maintenance report.
Review the output: does the agent identify the SSP rate discrepancy? Does it flag the missing FAQ entry? Does it recommend a link validity check?
Deliverable: A configured onboarding orchestrator with a traced T-14 to Day 90 workflow for Ayesha Raza, and a simulated monthly maintenance report from the policy maintenance agent with at least three findings and prioritised recommended actions.
Try With AI
Use these prompts in Cowork or your preferred AI assistant.
Reproduce: Trace the onboarding orchestrator workflow for Ayesha Raza.
I am testing an HR onboarding orchestrator agent. Please trace the full
workflow for the following new hire:
New hire: Ayesha Raza, Senior Data Analyst
Start date: 10 March 2026
Manager: Omar Farooq, Head of Analytics
Company: EdTech company, 250 employees, Karachi
Mandatory training: GDPR, Information Security, Code of Conduct
HR Business Partner: Aisha Butt
For each of the eight milestones (T-14, T-7, T-3, Day 1, Day 10,
Day 30, Day 60, Day 90):
- What does the agent check or action?
- What does it send and to whom?
- What would trigger an alert?
Also: at T-3, the pre-boarding checklist shows the laptop has not been
confirmed as ordered. What does the agent do?
What you are learning: Tracing the full workflow milestone by milestone makes the agent's value concrete. The T-3 laptop scenario illustrates the specific failure the orchestrator prevents — the one that ruins Day 1 for a new hire and takes weeks to recover from.
Adapt: Design the alert trigger conditions for your organisation.
I want to configure the alert trigger conditions for an onboarding
orchestrator at my organisation.
Context:
- Company: [Size, sector, jurisdiction]
- Typical new hire profile: [Role level, department, remote or on-site]
- Our current onboarding failure modes: [What actually goes wrong most often?]
- HR capacity: [How many HR team members handle onboarding?]
Please design three alert trigger conditions specific to our context.
For each:
- What event or non-event triggers the alert?
- What is the urgency (immediate / next business day / weekly report)?
- Who receives the alert (HR, manager, IT, other)?
- What action should the alert request?
Also: what is the single most common onboarding failure that agents
could catch that humans currently miss — based on our context?
What you are learning: Alert trigger conditions are the most important configuration decision for any event-triggered agent. An agent that alerts too frequently is ignored. An agent that only alerts on critical failures is genuinely useful. Designing the conditions for your specific context forces you to identify what actually goes wrong — not what theoretically could.
Apply: Design a monthly policy audit cycle for your organisation.
I want to design a monthly policy audit cycle for my organisation
using a policy maintenance agent.
Context:
- Company: [Size, sector, jurisdiction]
- Current policy documentation: [Where do policies live? How many?
When were they last reviewed?]
- Known issues: [Any policies you know are out of date or inconsistent?]
- Jurisdiction: [UK / Pakistan / other — for statutory rate monitoring]
Please design:
1. The five monthly checks the agent should run (use the standard five
categories but adapt them to my documentation set)
2. The event triggers for statutory rate monitoring in my jurisdiction
— what rates should the agent monitor and when do they change?
3. The output format for the monthly report — who receives it,
what decisions should it prompt, and what is the escalation path
if a critical issue is found?
4. What is the one policy most likely to be out of date in my
organisation right now? What should I check first?
What you are learning: A policy maintenance agent is only useful if it is configured for your specific documentation state and jurisdiction. Designing the audit cycle for your context reveals the gaps that exist right now — before you even deploy the agent. The final question ("which policy is most likely out of date right now?") often produces the most immediately actionable output.
Flashcards Study Aid
Continue to Lesson 13: People Analytics and Agent Operations →