AI in talent work
·
8 min read
Key findings
Korn Ferry found that 52 per cent of the 1,674 talent leaders it surveyed plan to add autonomous agents to teams in 2026.
Every agent needs a named person with authority to stop it, hours to check it and access to its record.
In a recruiter experiment, higher-quality AI led to less time per résumé and worse decisions than weaker AI or none.
HR decides the task split, role descriptions, review criteria and challenge route, while IT and risk own access, data security and testing.
A one-page agreement and a ninety-day review give the team a fixed point to widen, keep or withdraw the agent.
Every agent needs one named person who answers for its output, recorded in the team agreement.
When AI agents join a team, HR owns four decisions about people.
They are who answers for an agent's output, which tasks stay with people, how roles and reviews change, and what is recorded when a decision is challenged.
IT and risk own the technical controls.
Korn Ferry surveyed 1,674 global talent leaders and 230 of its own talent experts for its 2026 talent acquisition research.
Of the leaders, 52 per cent plan to add autonomous agents to their teams in 2026, and many are already creating employee records for AI agents in HR.
The same release names collaboration between agents and people, including onboarding hires and tracking performance, as a skill many leaders lack.
Gartner's 2026 survey of 426 chief human resources officers in 23 industries lists a talent strategy for a blended workforce among four priorities.
The page does not discuss agents themselves, so the practical questions fall to the HR adviser inside the business.
HR writes its decisions into a one-page agreement that the team signs before the agent starts.
What Current Guidance Leaves Out
Published advice tells HR to treat an agent as a new hire.
Give it a scope of work, key performance indicators and a manager, run quality checks, and train staff on what it can and cannot do.
The comparison helps with scoping and misleads on accountability, because an agent has no contract and cannot be disciplined or asked to explain itself.
Five questions stay open after the new-hire advice has been followed.
Who answers when an agent's output causes harm
How the work is audited, and by whom
How reviews of the people on the team change
How an employee challenges a decision the agent contributed to
What a manager must be able to do before supervising an agent
Each is a people decision, and each can be settled before the agent starts.
A Named Person For Each Agent
Every agent needs one named person who answers for its output, recorded in the team agreement.
The person is a human being with a job title, and a deputy is named for leave.
Enterprise guidance on agents lists clear ownership, with accountability and operational funding, among the conditions for deployment.
Three conditions make the name mean something.
Authority to stop the agent or reject its output without asking permission
Time in the workload to check its work, stated in hours per week
Access to the record of what the agent did
Without them the person is an accountability sink, a phrase of Dan Davies that Cory Doctorow applies to a radiologist placed between an AI and a patient.
The role lets the organisation deploy the machine without taking responsibility for its failures.
Time matters because people check less when the tool is good.
Mollick reports that Fabrizio Dell'Acqua gave 181 recruiters 44 applications to evaluate for mathematical ability, with good, poor or no AI support.
Recruiters with higher-quality AI spent less time and effort on each résumé, followed its recommendations without scrutiny, and made worse decisions than recruiters with weaker AI or none.
In a separate consulting experiment, one task was built to lie outside the AI's competence.
Consultants working alone solved it correctly 84 per cent of the time, and consultants using AI did so 60 to 70 per cent of the time.
What a talent function may let AI draft and decide sets out the same evidence and the line it supports.
What A Manager Must Explain
A manager who supervises an agent must be able to explain its work to an employee without the supplier present.
The manager needs five answers for any output.
The task the agent was given for this case
The information it used
What it produced
What the person checked and changed
What would have changed the decision
Kakade and Vispute state that employees and stakeholders must be able to understand how an AI system reaches a decision or recommendation.
They add that a lack of understanding breeds mistrust and weakens compliance with legal requirements.
A manager who cannot give these five answers is not ready to supervise the agent.
Tasks The Agent Takes
HR defines the work an agent takes by taking a role apart, task by task.
Mollick sorts tasks into three kinds.
Just Me tasks stay with a person, including decisions and tasks that carry values
Delegated tasks go to the AI and are checked by a person
Automated tasks are left to the AI without checking
He expects the automated group to be small for now, because AI still makes too many mistakes to run unattended.
Consider a talent operations team at a retailer, with a lead, three coordinators and an HR business partner.
The team lists every recurring task and assigns each one to a column.
Delegated, every item checked. Interview scheduling, onboarding checklists and first drafts of candidate status messages
Moved to sampled checks. Scheduling and checklists after thirty days, once the error log is clean
Kept by people. Shortlisting, offers, any message about pay, and any concern about performance or conduct
Excluded. Ranking or rating individual employees
The columns become part of the team agreement, and a task moves between them only by a recorded decision.
A manager who cannot give these five answers is not ready to supervise the agent.
Roles, Objectives And Reviews
Role descriptions change in three places.
The named person's description carries the supervision duty and the hours it takes
Each coordinator's description states which tasks moved and which judgements now take more of their time
The lead's description includes a quarterly review of the agreement
Objectives are set on the combined result of people and agent, such as time from offer to start date, and not on the volume the agent produces.
A review of the named person asks how often they found errors, which ones, and what they did about them.
A person who never overrides a capable agent may have stopped checking, which is the pattern in the recruiter experiment.
The agent is not rated on the employee scale.
It has its own measures, such as error rate, rework time and escalations, reviewed each quarter by the named person.
Rating discipline for the people is covered in performance ratings under pressure.
Closing The Manager Gap
Korn Ferry reports no percentage for the shortage of leaders who can work with agents, so HR has to measure it locally.
The main hazard is automation bias, the tendency to over-trust suggestions from an AI even when they are flawed.
The recruiter and consultant experiments above show it in trained professionals.
A manager qualifies by practice and not by attendance at a briefing.
Before an agent starts, each supervising manager reviews twenty outputs from the agent's real task, some containing errors that the HR adviser has planted.
The manager must find the errors and say what each would have cost.
The result sets the checking rate for the first month.
Supervision goes to people who have done the task themselves, because a manager who cannot perform the task cannot check it.
Kesby names four capabilities for leading human and AI teams.
Task allocation between people and AI
Prompting and curation of AI output
Performance management of a hybrid workforce
Ethical oversight, with humans kept in charge of critical decisions
A development plan for each supervising manager covers these four.
Records For Audit And Challenge
Two records are needed, and HR owns the design of both.
The register holds one entry for each agent.
Purpose and the tasks it takes
Named person and deputy
Information it may read, as agreed with IT
Date and description of every change to its instructions or model
Next review date
Whether the register sits beside personnel records or apart from them is settled between HR and the owner of the HR system before the first agent is entered.
The decision record is the second.
For any decision about a person that an agent contributed to, it states what the agent produced, who checked it, what the person changed and who decided.
The agreement gives an employee who challenges such a decision a route to a fresh review by someone who did not take the first one, with the record in hand.
Kakade and Vispute list accountability among the legal issues of AI in HR and call for clear mechanisms for assigning responsibility and addressing the consequences of AI-related mistakes.
Employment counsel confirms which explanation rights apply in each country, and HR does not guess.
The One-Page Agreement
A team adopting an agent signs a single sheet before the agent starts, using the structure in decision rights inside a team.
Purpose. The agent's job in one sentence
Named person. One person and a deputy, with weekly hours
Tasks taken. Each task by name, with the check applied to it
Tasks kept. The tasks and decisions reserved to people
Limits. What the agent may not do, including ranking or rating people
Stop right. Who can suspend the agent, and how quickly
Records. Where the register entry and decision records are kept
Challenge route. How an employee asks for a fresh review
Review date. Ninety days from the start, then quarterly
The team lead, the named person, the HR adviser and a contact in IT or risk sign it.
Technology And Risk Controls
Access rights, data security, model selection, testing, logging and incident response belong to IT and risk.
HR asks for three things and designs none of the controls.
The list of information the agent can read
Confirmation that its actions are logged
The name of the person to call when it fails
HR that designs technical controls takes on accountability for systems it cannot inspect.
First Ninety Days
Days 1 to 10. Sign the agreement, split the tasks and qualify the supervising managers
Days 11 to 30. Check every output, with the named person logging errors and hours
Days 31 to 60. Replace full checks with sampled checks on tasks that have a clean record
Days 61 to 90. Review the log with the team and revise the agreement
At day ninety the team decides, task by task, whether to widen, keep or withdraw the agent's work.
Start with one team that already runs an agent informally, and write its agreement this month.
Where The Tools Sit
Trusted Talent Advisor ships with The Prompt Library, which holds a prompt for each of the 51 tools and five more for the diagnostic lenses.
The Talent Template Pack includes the Talent Risk Register, which records an owner and a review date for every risk, the form an agent's risk entry can take.
Sources
1. Korn Ferry. Korn Ferry Research Unveils Top Talent Acquisition Trends Shaping 2026. Korn Ferry.
2. Gartner. Top HR Trends and CHRO Priorities for 2026. Gartner.
3. Mollick, E. Co-Intelligence: Living and Working with AI. Portfolio.
4. Kakade, K. S. and Vispute, S. A. AI Dynamics in HRM. Apple Academic Press and CRC Press.
5. Doctorow, C. The Reverse Centaur's Guide to Life After AI. Verso.
6. Kesby, M. Untangling AI: Driving Business Success Through Enterprise Automation and AI Agents. Wiley.
7. Milchanowski, K. L. Return on Intelligence: A Strategic Enterprise Playbook for Scalable AI Agents. Routledge.
8. Soofastaei, A. LIFE 4.0: Human Life in the Age of Artificial Intelligence. CRC Press.
9. O'Brien, C. How HR Leaders Can Handle Their Newest Colleagues: AI Agents. Raconteur.
10. Bade, F., Douet, N. and Chmara, R. From Tool-Mate to Teammate: How HR Has to Lead the Rise of Agentic AI. Deloitte Luxembourg.
From the system
Trusted Talent Advisor sets out the full method: seven components, 51 tools, The Talent Template Pack and The Prompt Library.
See the system
Questions this answers
How do you onboard AI agents into a team?
Write a one-page agreement that names the person who answers for the agent, splits tasks into those the agent takes and those people keep, and sets a review at day ninety. Qualify the supervising managers, and check every output for the first month.
Who is accountable when an AI agent makes a mistake?
A named person, recorded in the team agreement before the agent starts. That person needs authority to stop the agent, hours in the workload to check its work, and access to the record of what it did. Without these the name is a formality.
What is HR's role in managing AI agents?
HR owns the people decisions, which are accountability, the split of tasks between agent and person, role descriptions, reviews, manager capability, records and the route for an employee to challenge a decision. IT and risk own access, data security, testing and logging.
How do you manage AI agents as team members?
Give each agent a purpose, a named supervising person, measures such as error rate and rework time, and a review date. The agent is not rated on the employee scale. The people around it are reviewed on how well they check, challenge and override its work.
What should be recorded when an AI agent contributes to a decision about an employee?
The record states what the agent produced, who checked it, what the person changed and who decided. It sits beside a register entry that holds the agent's purpose, named person, permitted information and every change to its instructions.
How many companies plan to add AI agents to their teams?
In Korn Ferry's 2026 research, 52 per cent of the global talent leaders it surveyed planned to add autonomous agents to their teams in 2026. The survey covered 1,674 external leaders, and the release reports no percentage for the skills gap.
Keep reading
Measuring HR impact
Internal Mobility Metrics Without A Platform
Seven measures an HR adviser can build from existing records, each with a stated definition, a base and a way to read it.
7 min read
Workforce and future skills
Job Rotation As A Talent Gap Strategy
A job rotation programme closes a talent gap when the role, the person, the cover and the evidence are agreed with both managers before it starts.
8 min read
Talent reviews and succession
Seeing Talent Below The Top Team
Visibility of leadership potential drops below the top team. A cross-unit forum, run on one page per unit and a written evidence standard, is the method for seeing it.
8 min read