Ask The Analyst: What's the last HR workflow you'd trust to run fully autonomously with AI?

Each week for The Briefing, UNLEASH’s weekly intelligence email for senior business decision-makers, we ask the true experts – our community of analysts – to solve the biggest HR challenges.

Got a question for the analysts? You can send it directly to them via The Briefing – make sure you’re signed up.

This week’s question: What's the last HR workflow you'd trust to run fully autonomously with AI, and why?

Here are their perspectives.

Emily Rose McRae, Senior Director Analyst, Gartner:

For now, probably internal talent marketplace matching recommendations for both internal talent and hiring managers. Which is because it still has a human element for final review.

Honestly, none – not fully autonomously, and that’s by design rather than a technology gap we’re waiting to close, though I’m open to the possibility that the technology will change and this will be a different answer in the future.

The reason is structural: AI can share the work, but it can’t share accountability. Even agentic AI only acts within the scope of what humans have set or allowed, it lacks its own goals, self-reflection or genuine understanding of the task. That gap matters most on judgment calls with real consequences such as hiring decisions, performance actions or anything touching legal or reputational risk.

Organizations getting AI adoption right tend to be strategic about this – their question isn’t “where could AI do a task” but rather the “which AI opportunities are compatible with our strategic goals, risk tolerance, technology stack, and budget?” It’s a less exciting question, but far more strategic.

For all of the key workflows, the successful organizations are insuring there is at a minimum a human layer to evaluate and approve AI-generated outputs rather than letting systems act automatically, preserving the ability to catch and correct errors before they compound, which is particularly important for agents, because AI agents can drift and stop performing as intended.

Jean-Baptiste Audrerie, HR tech industry analyst, NexaRH:

Any process involving intent, judgment, and consequential decisions about people. That covers a vast range of complex tasks. The most relational and sensitive of all is probably recruiting.

The rule is simple: Digital, transactional, high-volume work with explicit rules and abundant data belongs to agents. Everything else belongs to humans. Humans design, decide, and stay accountable. Agentic AI excels when data is reliable and policies are explicit. But AI drifts when instructions are not explicit and incremental. And it produces more reports than any human can read.

In theory, recruiting could be automated end to end, from job requisition to offer signature. The technology is getting there. The real question is not "Can we ?" but "Should we?"

Some sanctuaries will remain AI-free by choice, and by necessity now that candidates use AI on their side. Is the candidate a real person? Is the employer brand a real, living culture? Yes, agents are extraordinary force multipliers, especially for frontline and high-volume recruiting. They answer candidates' questions in 20 languages, assess skills, detect fake resumes and bots, summarize interviews, schedule meetings with hiring managers in seconds, and personalize onboarding at scale.

AI can challenge staffing needs, candidate data, and recruiting insights. It should not make decisions automatically at every stage, especially where the human touch creates unique and decisive relational moments. The transactional layer is only 25% of the work. The other 75% is intangible: culture, emotions, informal networks, judgment. That reality never reaches an agent's context window.

The winners will treat AI as a brilliant digital sparring partner. The higher the collaboration, the better the outcome.

Michael Rochelle, Chief Strategy Officer & Principal Analyst, Brandon Hall Group:

The last HR workflow I'd trust to run fully autonomously is employee termination. While AI can streamline many aspects of offboarding including generating documentation, revoking system access, coordinating equipment returns, ensuring compliance checklists are completed and scheduling exit communications.

The decision itself should never be left entirely to a machine. Termination decisions carry significant legal, ethical, financial and human consequences. They often involve context that AI cannot fully understand: individual circumstances, performance history, accommodation requirements, organizational politics, cultural considerations and the potential for bias or unintended discrimination.

Even if AI can identify patterns that suggest a termination may be warranted, a human leader should always review the evidence, validate the recommendation and take responsibility for the final decision.

More broadly, the closer an HR process gets to materially affecting someone's livelihood, career, or dignity, the more important human judgment becomes. AI is exceptionally valuable as an advisor by surfacing risks, analyzing documentation, checking for policy consistency and ensuring compliance. Accountability for high-impact employment decisions should remain with people.

In short, I'd trust AI to automate the administrative work surrounding termination, but not the decision to end someone's employment. That remains one of the last HR workflows where human judgment, empathy and accountability are indispensable.

Kathi Enderes, SVP Research, The Josh Bersin Company:

Every HR workflow can run autonomously. None can run unaccountably. Sourcing, screening, pay adjustments, performance reviews, succession, learning.

Our HR 2030 research maps 158 agents across eight superagents, and there isn't a workflow on that list AI can't execute. In some companies, parts of it already do.

The real question is which one I'd last trust to run without anybody watching. Trust isn't earned by accuracy. A 99% accurate agent has no business running a termination unattended, because the 1% can't be undone. A cruder agent can run interview scheduling anytime (the worst case is a rescheduled meeting).

I'll let a workflow run fully autonomously when three things are true: the decision is reversible, the reasoning is auditable, and a named human owns the outcome.

Scheduling and sourcing clear that bar. Pay adjustments and terminations don't. So the last workflow I'd trust is any workflow where the decision can't be undone and the owner can't be named. That's not a category of HR work. It's a test you run on all of it. AI can run the workflow. It cannot own what happens to the person on the other side of it. That's what HR leadership means in the superworker era.