AI-Native People Team Blueprint

Source: core/clients/madison-reed/work/MR-people-ai-native-org-blueprint-2026-06-24.md Purpose: The clean-sheet design for what the MR People function looks like rebuilt from the ground up on an AI-native model. Internal confidential; the client-facing version is a scrubbed summary. Sensitivity: Contains personnel-level recommendations. Do not share directly with MR.

The core idea

If you rebuilt the People function today, agents would carry the processing and monitoring, and humans would own the judgment, the relationships, and the field work. No human processing tier underneath the senior owners.

This inverts the traditional model:

• Traditional: Director, managers, HRBPs (Human Resources Business Partners), plus a processing layer (administrators doing data entry, form routing, report generation)
• AI-native: Domain owners directing agent fleets that carry the processing. Humans staffed only to genuine judgment and field presence.

Current state vs. end state

Today	End state
Director + managers + HRBPs + processing tier (Roxy, Beth, two proposed backfills)	People lead, domain owners, one builder
Nine platforms in silos, one human adjudicating integrations	One HCM core + governed residual stack + agentic intelligence layer
Processing by people, automated "eventually"	Processing by agents first; humans staffed to residual judgment
Operators re-key data across systems by hand	Live connectors + semantic data layer; agents read, humans decide
Architecture owned by one saturated node (Shlanda)	Small frontier team ships under standing policy; governance board handles exceptions
Freed capacity re-hired as headcount	Freed capacity drops to P&L or funds higher-altitude judgment

The six domain pods

Each pod owns a real business outcome, not a task list. Each has one senior owner and an agent fleet.

Domain	Owner	Outcome	Key metrics
Talent Acquisition	Lindsey	Time-to-fill at quality (hourly + corporate)	Cost-per-hire, agency-fee avoidance, First Advantage cycle time, 90-day retention
Compliance, Total Rewards & Benefits	Grecia	Compliance risk closed; wage/premium exposure down	Workers' comp mod factor, wage-violation rate, benefits-admin cycle time, comp-benchmark coverage
HRBP / Field	Shantae, Amy	Engagement and bottom-quintile HCB same-store lift	ER case throughput, manager-coaching reach, voluntary turnover, field-presence days
Employee Lifecycle	Syra	Day-one readiness and clean transitions	Onboarding cycle time, job-change/term error rate, offboarding completeness
L&D	Nickole	Org AI fluency and capability	Fluency-ladder movement, training-to-competence time
People Systems	The builder (your role)	Agent fleet uptime and reuse	Agents in production, eval pass rate, integration coverage, human-hours returned

Agent fleets per domain

Compliance, Total Rewards & Benefits (the flagship, already partially live)

• Leave-of-absence agent (LIVE, Phase 2)
• Pay reconciliation agent (LIVE, Phase 2)
• Minimum wage monitor (LIVE, Phase 2)
• Benefits-administration agent (planned)
• Comp-benchmarking agent (planned)
• Compliance and wage-audit agent (planned)

Talent Acquisition

• Sourcing agent (hourly + corporate funnels)
• Screening and scheduling agent
• First Advantage / I-9 orchestration agent
• Offer-and-onboarding-handoff agent
• Pipeline watchdog (time-to-fill and cost tracking)

HRBP / Field

• HR inbox and Slack deflection bot (Tier-1 policy Q&A; most-cited time sink at 12+ hrs/week for Shlanda)
• ER-case (Employee Relations) triage agent
• Engagement-pulse synthesizer (Culture Compass survey signals turned into manager-actionable summaries)
• Bottom-quintile HCB diagnostic (surface which locations need a field visit and why)

Employee Lifecycle

• Onboarding day-one orchestrator (the 9-system handoff)
• Job-change and termination workflow agent (the 100+ task checklist, structured)
• Offboarding agent

L&D

• Training-content synthesis agent (Nickole already uses Claude for this)
• Survey-to-insight agent
• AI-fluency curriculum delivery

The People brain

The blueprint calls for a central knowledge repository (Git + RAG) that every agent reasons from:

mr-people-brain/
  codebook/                     # shared People brain
    policy/                     # handbook, PTO, conduct (source of truth)
    compliance/                 # multi-state wage rules, FMLA/ADA, premium-pay
    total-rewards/              # comp philosophy, job architecture, market bands
    playbooks/                  # onboarding, job-change, termination, ER runbooks
    personas/                   # role and population definitions
    decisions/                  # decision log: every exception, every ruling, why
  domains/
    talent-acquisition/         # agents, queries, runbooks, KPIs per domain
    compliance-tr/
    hrbp-field/
    lifecycle/
    l-and-d/
  platform/
    connectors/                 # Paylocity/Workday, Legion, Greenhouse, etc.
    orchestration/              # schedules, event triggers, run log
    evals/                      # eval suites per agent
    governance/                 # accounts, permissions, HITL gates, risk tiers

Everything in codebook/ is embedded and retrieved, so every agent answers from the same policy, wage rules, and comp philosophy. The decision log sharpens the brain each cycle.

The data layer

Three tiers:

1. System of record: Paylocity today, Workday in 2027. Canonical for employment, pay, leave, benefits, position.
2. Residual stack: Specialized truth that the core HCM can't replace (Legion for retail scheduling, HR Acuity for investigations, First Advantage for background checks, Guardian for insurance).
3. People semantic layer: Core metrics (time-to-fill, mod factor, wage compliance, turnover, admin-load ratio, engagement) defined once as governed models. Agents query reliable numbers instead of re-deriving them from raw data every time.

Write access rule: Agents read with least-privilege scopes. They do not write to employee records. The LOA agent proposes a Paylocity re-entry and a human approves. The wage agent flags exposure and routes it. Write access is human-in-the-loop, especially for pay, leave status, and compliance.

CMPRSSN's role and hand-off plan

1. Now (Phase 2): Finish the Compliance pod (leave, pay recon, minimum wage agents; the existence proof)
2. Next: Stand up the People brain as a Git + RAG repo; define the semantic layer on Paylocity (portable to Workday)
3. Then: Runtime and governance scaffolding (evals, least-privilege scopes, human-in-the-loop gates, risk tiers)
4. Then: Fleet build-out (TA, HRBP/Field, Lifecycle domains, in priority order)
5. Throughout: Scaffold the builder role. The People Systems builder is the seat CMPRSSN hands the platform to. CMPRSSN operates midterm and transfers ownership as the builder ramps.

The builder role is your role. The end state is that MR owns everything: the brain, the agents, the governance, the platform. CMPRSSN's success condition is a function that runs without CMPRSSN.

Why the timing matters

Two windows open and close together:

• Contract decisions (Greenhouse renewal, Workday evaluation) force stack re-choice this year. This is the cheapest moment to make agentic-first choices.
• Two hires are being defined right now. Re-scoping the People Systems JD from administrator to builder is a zero-cost edit now and an expensive miss later. Once the hire is made at administrator altitude, the agentic layer has no owner.

What this means for your embed

You are executing steps 1-5 simultaneously:

• Month 1 (August): You're in the seat doing the manual work (finishing the Compliance pod's existence proof, while instrumenting the Lifecycle workflows)
• Month 2 (September): You're building agents and cutting over (standing up the brain, building the lifecycle fleet)
• Month 3 (October): You're handing off (the team runs what you built through peak season, proving it works)

The end-state blueprint is the destination. Your three-month embed is the bridge from where things are today to the team owning this themselves.