Manage people and AI agents by assigning work at the task or workflow level, setting clear limits on what each agent can do, and keeping a named person accountable for consequential outcomes. An agent’s ability to perform a task is only one consideration: worker preferences, error consequences, data access, handoffs, and human judgment matter too. That is a practical management approach—not proof that every workplace will use autonomous agents or that agents have the status of employees.
Why managing agents is a management problem, not just a technology decision
When an AI agent can complete part of a job, leaders still have to decide whether it should, how much authority it gets, and who is responsible for its work. Harvard Business School’s Managing the Future of Work page summarizes Joseph Fuller’s March 2026 guidance as applying familiar human-resources ideas to agents: define roles and boundaries, assign supervision and accountability, and review performance. Those ideas offer a useful starting point, but they do not make an agent a person or settle its legal or moral status.
As an Amazon Associate I earn from qualifying purchases.
The central question is not simply “Can the agent do this?” It is also “Should this task be automated, shared, or kept under human control?” Those questions can have different answers even within the same workflow.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →How to decide which work belongs to people, agents, or both
Start with discrete tasks rather than assigning an entire job to an agent. A job may combine routine processing, decisions that require context, customer or coworker interactions, and actions with very different consequences. Consider each task in its actual workflow and use these questions as a decision aid—not as a validated scoring formula:
#1 Best Overall
- Capability: Is the agent estimated to be capable of doing this task, and under what conditions?
- Worker preference and agency: Do the people doing or affected by the work want automation, shared work, or a human decision-maker?
- Impact of error: What could go wrong, how serious would the harm be, and who would notice?
- Reversibility: Can an action be checked and undone before it causes lasting effects?
- Access: What sensitive information, systems, or permissions would the agent need?
- Ownership and handoffs: Is a human clearly responsible, and can the agent pass work to that person with enough context?
- Measurement: Can the team assess the full workflow, including corrections and escalation, rather than counting only the agent’s completed actions?
Stanford’s WORKBank project offers a way to keep capability and preference distinct. Its four zones—Automation Green Light, Automation Red Light, R&D Opportunity, and Low Priority—combine workers’ desire for automation with assessments of agent capability. The project’s summary does not make those categories a universal decision rule; they are a framework for asking both whether an agent may be able to do a task and whether workers want it automated.
For example, an agent might draft or organize information while a person decides what to communicate or authorizes an action. That division is a possible workflow design, not a claim that every task in those categories is suitable for automation. Set the boundary according to the particular task, its risks, and the people affected.
What an agent’s operating boundaries should make explicit
Before an agent is put to work, write down an operating agreement the team can understand and enforce. This is a practical recommendation synthesized from HBS’s agent-onboarding frame and the OECD’s attention to governance, not a universally proven formula.
- Purpose: Name the specific task or workflow the agent supports. Avoid a vague mandate such as “handle customer operations.”
- Authority: Specify which information it may use and which systems or actions it may access. Distinguish work it may prepare from work it may execute.
- Human owner: Name the person responsible for monitoring the workflow and for its consequential outcomes. An agent’s output does not itself establish accountability.
- Escalation: Define when the agent must stop and ask for human review, and how it should hand over the relevant context. Include cases the agent cannot resolve confidently or within its authorized scope.
- Review: Set a process for checking quality, errors, exceptions, and whether the agent is still serving its defined purpose.
These boundaries should be understandable to the people expected to work with the agent. If the team cannot tell what the agent is allowed to do, when it must stop, or who takes over, the workflow is not yet clearly managed.
How to evaluate the people-and-agent workflow
Review the combined work, not just the agent’s output. A workflow can appear fast while shifting hidden effort to employees who must check, correct, explain, or redo its work. A useful review should make those handoffs visible and assess whether people can intervene when needed.
- Check the quality of completed work and the handling of mistakes or exceptions.
- Account for human review, correction, and escalation when assessing the workflow.
- Confirm that the agent stays within its permissions and that responsibility for decisions remains legible.
- Ask the affected workers whether the division of work supports appropriate human agency and workable collaboration.
HBS’s discussion explicitly brings accountability and performance reviews into the management of agents. Microsoft’s 2026 Work Trend Index likewise treats readiness as a system involving leadership, culture, management practices, and measurement—not merely access to a tool. A review should therefore consider how the workflow operates for the team, rather than treating agent deployment as a stand-alone technical event.
What current evidence says—and what it does not establish
Different kinds of evidence answer different questions. Worker preference studies can help identify what people want automated; capability assessments estimate what agents may be able to do; product telemetry records activity in a particular environment; interviews describe organizations’ experiences. None, on its own, establishes a universal productivity effect or predicts what will happen in every workplace.
Free tools Windows power users keep installed
One-click scans. No signup required.
- Stanford WORKBank: The 2026 project summary describes worker preferences and agent-capability assessments covering 1,500 workers across 104 occupations and 844 tasks. It reports a mismatch between some AI-development priorities and workers’ expressed task priorities; many tasks where workers see a role for equal human-agent collaboration while generally preferring greater human agency; and a possible shift in valued competencies from information-processing toward interpersonal and organizational skills. These are findings from a worker-centered U.S. audit, not a forecast for all countries, employers, or occupations.
- WORKBank’s task framework: The associated preprint, version 3 last revised February 1, 2026, describes the project as using U.S. Department of Labor O*NET tasks, worker preferences across 844 tasks and 104 occupations, and AI-expert capability assessments. The project summary reports that 41.0% of the Y Combinator company-task mappings it analyzed fell in the Low Priority and Automation Red Light zones. That percentage applies to those mappings, not to all AI companies or the labor market.
- Microsoft’s 2026 Work Trend Index: Its methodology says the North American commercial user-goal mapping, excluding education, used 105,000 samples from one week in February 2026. That is a sample count, not an estimate of the worker population, and the report draws partly on anonymized Microsoft 365 signals rather than a representative census of all work. Its People Science Agentic Teaming & Trust Survey was fielded in July 2025 and included 1,800 employees external to Microsoft: 819 leaders, 520 managers, and 461 individual contributors.
- OECD: The September 16, 2026 working paper is early evidence based on interviews with organizations across regions and sectors. It examines applications, benefits, challenges, and governance approaches; interviews should not be treated as a universal causal estimate of productivity or job displacement.
- National Academies: The 2025 consensus report assesses AI’s implications for U.S. productivity, the workforce, and education. Its official summary says AI “has the potential to enhance human labor and create new forms of valuable work – but this is not an inevitable outcome.” Potential is not a guarantee of job creation, productivity gains, or protection from displacement.
These sources support careful task-level decisions and ongoing evaluation; they do not justify a deterministic prediction about net jobs or a claim that every organization will benefit in the same way.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How a manager’s work may change
If routine information-processing tasks are delegated, managers may need to spend more attention on delegation, coordination, coaching, and judgment. Stanford’s WORKBank summary describes a possible shift in the competencies workers value, from information-processing toward interpersonal and organizational skills; it is a possibility, not a settled outcome.
Rank #4
- Stay present in every scenario: Every conversation is covered, in person, on calls, and online. 4 MEMS + 1 VPU microphones with AI beamforming capture every voice across the room. Smart Dual-Mode Recording switches automatically between phone calls and in-person. The free Plaud Desktop captures online meetings without a bot
- Walk out of every meeting with notes ready to act on: Plaud Intelligence transcribes in 112 languages with speaker labels and turns each recording into action items, decisions, and follow-ups, structured and ready to use. Choose from 10,000+ customizable templates tailored to your role and industry
- AI summary ready before you reach your desk: Auto Transfer moves each recording to the Plaud app automatically, and AutoFlow transcribes and summarizes so your notes are ready before you are back at your desk. Upgrade anytime to Pro (1,200 min/mo) or Unlimited
- Access your AI workspace anywhere: One connected workspace across Plaud Desktop, Plaud Web, and the Plaud mobile app, so your conversations and finished work follow you everywhere
- Your conversations stay private and yours: Compliant with ISO 27001, ISO 27701, SOC 2, HIPAA, GDPR, and EN 18031, with zero data used to train AI models. Trusted by 2.5M+ professionals, including legal, medical, and business professionals handling sensitive information
Team collaboration is also an open challenge. Microsoft Research’s New Future of Work says, “Collaboration between people is a central mechanism in how work gets done, but AI doesn’t yet work as well for teams as it does for individuals.” That observation cautions against assuming that an agent that helps one person will automatically coordinate well across a team. Managers should pay attention to shared context, handoffs, and whether people can understand and challenge the agent’s contribution.
The National Academies’ conclusion is similarly conditional: AI may enhance labor and create valuable work, but those outcomes are not inevitable. A manager’s job is therefore not to assume a predetermined future, but to make work allocation, oversight, and accountability clear as capabilities and workflows change.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
