Non-Human Resources: Why Your AI Workforce Needs Its Own Department
Every business has a department that manages its people. It hires them, onboards them, sets their permissions, tracks their performance, pays them and, eventually, manages their exit.
Very soon, many businesses will have more AI agents than people. Nvidia's Jensen Huang expects his company to have "hundreds of thousands of digital employees" alongside its 42,000 people (CNBC). Gartner predicts the average Fortune 500 company will have more than 150,000 agents by 2028, up from fewer than 15 in 2025 (Gartner).
So who manages them?
At the moment, in most businesses, nobody. Agents get built by enthusiastic individuals, connected to systems with whatever permissions were easiest, and left running long after anyone remembers why. Gartner calls this agent sprawl and found only 13% of organisations believe they have the right agent governance in place (Gartner).
In why most businesses are using AI wrong we argued that AI works best when it's organised like a team. We think managing that team needs a new business function: Non-Human Resources (NHR).
What is Non-Human Resources?
Non-Human Resources is the function responsible for the full lifecycle of a business's non-human workforce: AI agents, bots, automations and the models behind them. It does for agents what HR does for people.
Some HR leaders are already arguing along these lines. As one HR platform puts it, agents "need to be onboarded… They need a manager and a mentor… They need KPIs, and their performance needs to be monitored" (isolved). Security teams use similar language, treating every agent as a "non-human identity" with its own owner and access rights (Gartner).
NHR brings those ideas together into one accountable function.
HR vs NHR: the same lifecycle, a different workforce
| HR function | NHR equivalent |
|---|---|
| Workforce planning | Deciding which roles agents fill, which stay human and how they work together |
| Recruitment | Choosing whether to buy, build or configure an agent, and which platform and model to use |
| Job description | An agent role card: purpose, tasks, inputs, outputs, limits, escalation rules |
| Onboarding | Creating the agent's identity, granting least-privilege access, connecting approved data sources |
| Training and induction | Giving it context: brand guidelines, policies, knowledge bases, worked examples |
| Line management | Assigning a named human owner and a lead agent it reports to |
| Payroll and benefits | Token budgets, licences and compute costs, with spending caps |
| Performance reviews | KPIs, quality checks, error rates, value delivered against cost |
| Policies and conduct | Acceptable-use rules, guardrails, approval gates, data-handling rules |
| Compliance and records | Audit trails of every action, decision and data access |
| Disciplinary | Pausing, restricting or retraining agents that go beyond their scope |
| Offboarding | Retiring agents, revoking access, archiving their outputs and logs |
The five pillars of NHR
1. Identity and access Every agent gets its own identity, a named owner and only the permissions it needs. Microsoft has made this the default: since mid-2026, every new Copilot Studio agent automatically gets a Microsoft Entra Agent ID, which supports Conditional Access, audit logging and lifecycle management (Microsoft Learn).
2. The agent register You can't manage what you can't see. NHR keeps a central inventory of every agent, including the "shadow AI" agents built outside official channels. Gartner's first steps for managing sprawl are clear policies and a central agent inventory (Gartner).
3. Information governance Agents are only as trustworthy as the data they use. NHR decides which data each agent can reach, keeps that data current and stops oversharing. For Cyber Samurai, this is where data governance and AI governance meet.
4. Performance and cost Every agent has a job to do and a cost. NHR measures both. Token budgets are becoming a real line item: Huang has said he'd be "deeply alarmed" if a $500,000 engineer didn't use at least $250,000 of tokens (Yahoo Finance). Spending on that scale needs the same scrutiny as payroll, and agents that cost more than they deliver get reworked or retired.
5. Monitoring and conduct NHR watches what agents actually do, spots unusual behaviour and steps in when an agent goes beyond its remit. Gartner predicts that by 2029 at least 70% of organisations with production agentic AI in infrastructure and operations will have a material service, security or cost incident linked partly to weak runtime controls (Gartner). Data observability principles carry over directly.
Why HR or IT can't simply absorb this
It's tempting to give agents to IT ("it's software") or to HR ("they're digital workers"). Neither fits on its own.
- IT understands identities, access and infrastructure, but doesn't usually own job design, performance or workforce planning.
- HR understands roles, performance and organisation design, but not tokens, permissions or model behaviour.
- Business departments know what they need agents to do, but shouldn't each be making their own governance rules.
NHR sits between all three. It sets the standards, keeps the register and makes sure every agent has an owner, a purpose, a budget and an exit plan, while the departments actually manage their agent teams day to day.
Who does NHR?
In a large enterprise, NHR might become a team of its own led by a Head of Non-Human Resources. In an SME or mid-market business, it usually starts as a responsibility shared between operations, IT and data, often with outside help.
Grok Bot gives a sense of how the work is changing. Its bots "learn how you like work done" and can save a workflow as a routine and run it on their own (xAI). When agents learn and act like employees, someone has to manage them like a workforce.
Getting started with NHR
- Find what you already have. List every agent, bot, automation and AI tool in use, official or not.
- Assign an owner to each one, and retire anything nobody owns.
- Write role cards for the agents you keep.
- Tighten access to least privilege and give every agent its own identity.
- Set budgets and KPIs, and review them monthly.
- Define the lifecycle: how agents are requested, approved, onboarded, reviewed and retired.
Your people have HR. Your agents will need NHR. If you're starting to think about how to manage AI agents in your business, we'd be happy to talk it through.
Keep reading
