AI agents at work: what HR needs to plan for
Benjamin Manning, a PhD candidate at MIT Sloan, studies AI agents that act on people's behalf - and what that means for markets, institutions, and the way we make decisions. For HR, his work points to a near future where digital agents negotiate, schedule, draft, purchase, and even respond to surveys. The question isn't if this will touch your org, but how you'll make it safe, useful, and fair.
Who he is - and why his research matters to HR
Manning brings a mix of economics, policy, and computer science to a very practical problem: how to design and evaluate AI agents that make choices for people. He's also exploring AI systems that simulate human responses to help researchers and operators test ideas faster. This combination maps directly to HR work: decision rights, governance, policy pilots, and workforce upskilling.
Designing agents that act for employees
As more online actions are handled by AI, you'll need clear answers to basic questions: What can an agent decide alone? When must a human review? How do we record preferences so the agent represents people accurately? HR can lead by setting guardrails that protect employees while letting teams move faster.
- Define decision rights by risk tier (e.g., the agent can schedule meetings and draft job posts; a manager reviews offers and compensation).
- Create a simple, employee-owned preference profile (communication style, working hours, budget limits, values constraints).
- Require human checkpoints for material decisions and any action that affects pay, promotion, or termination.
- Log agent actions for audit, with opt-out controls where legally required.
Agents will influence how markets and institutions work
Manning's research looks at how agent behavior changes outcomes, not just individual tasks. For HR, that means vendor negotiations led by bots, candidate interactions with recruiters' assistants, and internal processes mediated by agents. Prepare for multi-agent workflows, incentives that apply to software helpers, and contracts that specify what an agent can or cannot do.
Simulated respondents: faster policy and program testing
Manning is also studying how well AI can simulate human responses. The practical use for HR is straightforward: pre-test policy changes, survey wording, benefits options, or training formats with synthetic populations before a live pilot. You cut false starts, then confirm with small human studies.
- Run synthetic A/B tests to narrow options (e.g., three PTO policies down to one or two).
- Validate with a small human pilot, then scale if results hold.
- Document gaps where simulations perform poorly (e.g., niche roles or regions) and adjust.
Guardrails to put in place now
- Adopt an AI governance standard consistent with the NIST AI Risk Management Framework and the OECD AI Principles.
- Set clear consent, transparency, and data minimization rules for any agent acting for employees or candidates.
- Establish bias testing and performance monitoring for agent-driven workflows (recruiting, scheduling, procurement).
- Create an escalation matrix for errors or harm, with rapid rollback options.
90-day action plan for HR
- Inventory repeatable decisions that could be delegated safely (calendar, FAQ replies, interview scheduling, benefits Q&A).
- Draft decision rights, thresholds, and review points by process and risk level.
- Pilot one agent-assisted workflow and one simulated survey project; define success metrics (quality, time saved, employee sentiment, fairness).
- Stand up basic audit logs and plain-language disclosures for employees and candidates.
- Report outcomes to leadership with a go/no-go recommendation for wider rollout.
Skills HR teams will need
- Prompt design and evaluation for day-to-day agent use.
- Experiment design and survey basics to run small, fast tests.
- Vendor assessment using model cards, privacy terms, and bias reports.
- Change management and communications that set expectations and reduce fear.
If your team is building these muscles, you can explore role-based upskilling paths and certifications here: Courses by job and Popular certifications.
The takeaway
Manning's work points to a practical shift: AI agents will take on more online activity, and AI-driven simulations can help you test ideas before you spend real budget or time. Organizations that set smart guardrails, keep humans in control for high-stakes calls, and build a simple test-and-learn habit will move faster with less risk. As he puts it, the pace of insight can finally catch up with economic change - if we do the work now.
Your membership also unlocks:
