Your Next Direct Report Might Be an AI Agent

Q Hamirani learned a surprising lesson when he asked AI to help with a compensation analysis: It completed the task without checking the decisions he wanted to review along the way. Hamirani, who serves as Chief People Officer at HighLevel, had to change the process so the system paused and confirmed job leveling, benchmark data, and final recommendations.

"I let it do too much almost end to end," he said during an appearance on "The Human Element."

As companies give AI agents responsibilities that used to sit with employees, we see the systems for assigning work, checking quality, and maintaining accountability aren't always keeping up. In BCG's 2026 AI at Work survey, 30% of respondents said their organization had integrated AI agents into workflows, up from just 13% in 2025. Another 50% reported agent experiments or pilots. Yet half of respondents said their companies lacked clear governance for teams that mix people and AI.

Meanwhile, some HR leaders have begun referring to AI agents as "digital teammates," or even adding them to org charts. As AI agents become so entrenched in our teams, what work are they supposed to do? What can they decide on their own? Who checks their work? And what happens to the people whose jobs they're changing? We asked HR leaders these questions and more — here's what they had to say. 

What Makes an AI Agent a Digital Teammate?

An AI agent is software that can work toward a goal, using information and connected tools to carry out tasks. Some agents only retrieve information or prepare recommendations, while others can update records, send messages, or initiate a sequence of actions with varying levels of human approval.

"Digital teammate" isn't a separate technical category; it's a way of describing an agent that's been assigned a continuing role in a team's work. The implication is that someone defines a role, gives the agent appropriate access, monitors its performance, and takes responsibility for what it does. Unlike a human teammate, though, the agent has no employment status (and, crucially, can't bear any legal responsibility for its own outcomes).

Before we move on, there's an important distinction to make: Vendors and those who talk about AI often use "agent" to describe a wide range of capabilities. A chatbot can answer questions, but usually can't take any actions. Generative AI can answer questions and summarize notes from meetings. An agent, or agentic AI, can open a case in an HR system or move a case forward. When discussing AI agents, it's important to know what actions they can complete, under what conditions, and with whose approval. 

But when it comes to AI agents that act as digital teammates, the most important distinction is whether the AI has an ongoing job that requires supervision. Many companies are experimenting with these types of agents; for example, Azurée S. Montoute-Lewis, Global Chief People Officer at Burson, described building 9,000 internal agents for the 5,000-person company. One helps managers and employees prepare for difficult feedback conversations using the firm's feedback model. Another is based on Burson's learning and development initiatives and coaches employees to help them grow.

Who Manages an AI Teammate?

Every agent ultimately needs human accountability.

"There always needs to be a human owner of the outcome," Hamirani said. Gartner makes the same distinction, stating that AI agents can act, but they aren't legal entities that can be held responsible for the results of their actions.

In practice, there may not even be just one person responsible for managing AI teammates; accountability may be distributed across multiple leaders or teams. For example:

• A business owner who defines the agent's purpose, outcomes it should produce, and what counts as successful or acceptable performance
• A technical owner who manages integrations, access, permissions, security, and changes. This person may also have the ability to shut the agent down if needed.
• A human decision-maker or reviewer who approves (or denies) actions that require professional judgment. This person may also give explicit authorization for sensitive decisions, such as pay changes or resolution to employee relations cases.

One person can hold more than one of these roles; the more important thing is that none of them is left empty. HR can naturally fit into most, if not all of them, or can work with IT, security, legal, and business stakeholders to cover all bases.

What Are Best Practices for Managing Digital Teammates?

Once ownership is clear, managers need a way to put it into practice. Before an agent gets access to company systems, implement these three practices to turn its responsibilities into work your team can trust.

Give Agents Job Descriptions Before Giving Them Access

Hamirani described bringing an agent onto a team as closer to onboarding than installing software. While he said he doesn't necessarily write a traditional job description, he does start with the same question each time: "What am I expecting AI to help me with?" The documentation becomes, as he put it, "almost 2.0 of an SOP or a standard operating procedure."

So while a job description — at least, a traditional one — isn't strictly necessary, an AI agent needs a documented assignment, not just a prompt. Here's what a role brief might look like for an HR agent:

This is just one example; a real role brief should reflect the AI product's realistic capabilities, what your organization has authorized it to do, and the risks associated with each workflow.

Set Agent Autonomy According To Risk

An agent that summarizes documents requires different control than one that can autonomously change an employee's compensation. Giving both the same permission to act — or subjecting both to the same approval process — just doesn't make sense.

Gartner has described four levels of AI agent autonomy:

More autonomy doesn't automatically make an agent more useful. HR work can be high stakes and involve a lot of steps, and not all of them belong at the same level. For example, an agent might gather salary benchmarks without intervention but need a person to validate the job level, review the analysis, and approve any resulting pay decisions.

That was, ultimately, Hamirani's solution to his compensation exercise: build checkpoints into the process rather than accept one answer at the end. His broader advice is to expand an agent's responsibilities in stages, with human checks placed at all points where judgment matters.

Keep Checking Agents' Work (Even After Launch)

Employees typically get coaching, performance reviews, and clearer expectations when their work isn't up to snuff. Agents also need ongoing evaluation, although the mechanisms are different: testing, reviewing outputs, adjusting instructions or configurations, and fixing problems in the data or integrations.

Christy Harris, SVP and CHRO at CCC Intelligent Solutions, spotted an interesting overlap. As leaders have begun managing people alongside AI systems, she questioned whether the precision needed to direct agents well could also help managers communicate more clearly with people. On "The Human Element," host Barb Bidan raised the idea that a manager who learns to improve vague AI instructions might also learn to give better feedback to employees.

"In the nature of having to continue to refine your prompts, to get really clear on what your feedback is," Harris agreed. "Is having to provide instruction, feedback, and clarity to an agent going to teach us that sort of clarity in other circumstances, as well as with the humans we work with?"

It's a promising thought. But it's still important to keep in mind that prompting an agent to try again isn't the same thing as improving its reliability. Managers need to know whether the underlying workflow is still producing correct, consistent results over time, which means looking at:

• Quality: How often does it give an accurate answer or complete the intended task, based on checks against a reliable standard?
• Escalations: Does it hand off the right cases, with sufficient context, before taking actions beyond its authority?
• Exceptions and incidents: What errors, policy conflicts, privacy concerns, or unauthorized actions have appeared?
• Effect on employees and HR: Are requests resolved appropriately, are employees getting help, and is the agent actually reducing work rather than creating more review and rework?

Should AI Agents Appear in the Org Chart?

They can, but a box labeled with the agent's name doesn't answer bigger-picture questions about managing AI teammates.

Conventional org charts show hierarchies of people, and agents sometimes do need to be mapped to workflows, systems, permissions, and human decisions-makers. For many organizations, it may be more useful to create a separate register for agents and link it to relevant teams or business processes. Then, use it to record agents' purpose, level of autonomy, access, checkpoints, owners, and other important information.

HR doesn't necessarily have to own the entire register, but it can help decide what information should be included and create a template for other teams to use when documenting their AI teammates.

Whether the org chart changes or not, the more consequential change comes behind it: Someone has to define and document the work, decide what stays human, and make sure employees have a future in the team being built. That's how HR can help shape the future of work as AI agents become more commonplace.

What Happens To Employees Whose Work Agents Take Over?

The org chart conversation gets a little more complicated when it moves from software to real people's careers.

Bradford Wilkins, VP of People and Organization at Cognite, recalled a manager who worried that AI would stop his team from growing, including taking away promotion opportunities that come from leading large teams. But he offered another possibility: "There might be management roles. They're just not management roles of people."

What Wilkins meant is that supervising agents may become increasingly valuable work. However, betting on it isn't a substitute for a career plan. And for people whose actual tasks disappear, simply being told to "do more strategic work" may fall flat, unless your organization is actively prepared to say what that work is and help them prepare for it.

Wilkins acknowledged the discomfort, saying that introducing AI into an established role can feel like "taking big parts of their job and chunking it away." He added that employers need to show people what their role could become and how the company will help them get there.

HR has a huge role to play in that. Work through changes with both managers and employees:

1. Map tasks that may move to agents. Clearly separate the work your organization plans to automate from the responsibilities and decisions that need to stay with people.
2. Redesign both roles and measures of success. Performance expectations should reflect the real work employees do, as that shifts over time.
3. Create modern development paths. Teach employees skills that are useful in the AI era, like how to evaluate AI output, oversee automated workflows, and exercise judgment agents can't provide. Be clear about what skills lead to advancement.
4. Involve employees in the redesign. The people doing the work know best about where an automated handoff is likely to fail, where a policy is ambiguous or unclear, and what customers actually need.

These concerns aren't just forward-looking or far off. In BCG's 2026 survey, 72% of respondents said AI had changed expectations for the skills they needed, but only 36% thought they'd received adequate upskilling.

Human Or AI, Your Next Report Still Needs a Manager

As AI agents become more common — and move into team roles — it'll likely shift how management works. For HR leaders, that means changing how work is assigned and where humans step in for the people on the team around the agent. It also means managing changing responsibilities — employees need clear expectations, support to build new skills, and a plan for their careers as we continue to move into the AI era.

Whether your next new hire is a human or an agent, make sure their manager knows what success in the role looks like, which decisions need human judgment, and how the team should respond when something goes wrong. An agent may be a new kind of direct report, but setting it up to succeed is still a human responsibility.