We talk a lot about what AI can do. Less about who owns it.

We'd spent much of the evening discussing AI use cases, and there were some genuinely interesting examples.

One HR team had built an agent to support its disciplinary process in one market, making sure steps weren't missed and records stayed up to date. Another had used AI to help develop a five-year people strategy and, in doing so, identified a labour compliance gap that hadn't previously been picked up. Then someone mentioned a global organisation that has created a leadership role responsible for both its people and its AI agents.

That changed the conversation and, for once, I don't think we could blame the wine. Because once an AI agent starts doing work that previously sat with a person, the question isn't just whether the technology works.

Who owns the outcome?

If the disciplinary agent misses a step, who is accountable?

If an AI screening tool consistently filters out the wrong candidates, who spots it?

If an agent gives incorrect advice on employment law or overtime, who is responsible for correcting it?

The room didn't really have an answer.

And I suspect most organisations don't either.

Is this actually HR's job?

Before the roundtable, we asked the leaders attending a few questions. Only 33% felt AI upskilling should sit with HR. Most felt responsibility should sit with line managers or IT. When it came to job redesign, 83% felt HR should own it jointly with technology.

On paper, that makes sense. AI has largely entered organisations through technology budgets, transformation programmes and vendor conversations. It therefore naturally feels like something IT should lead. The problem is that I've been in enough workforce conversations over the years to know that “joint ownership” can very quickly become no real ownership.

Everyone is involved. Everyone has a view. Nobody is ultimately accountable. One comment during the evening captured the problem particularly well. The C-suite is looking for productivity and cost efficiencies. Managers want better tools and faster execution. Employees are understandably thinking about what all of this means for their jobs and careers.

Those priorities aren't always going to line up neatly. Someone has to connect them and that isn't going to be the technology vendor.

The moment AI starts doing work, this becomes a workforce issue

There's an important distinction here. Using ChatGPT to improve an email is a tool. An AI agent screening candidates, supporting a disciplinary process, monitoring overtime against labour regulations or completing part of someone's workflow is different, it's doing work. Work that previously belonged somewhere in an organisation chart or someone's job description.

And that's where I think the conversation needs to move beyond AI adoption and into workforce design.

We heard plenty of examples in Bangkok of AI getting things confidently wrong: incorrect calculations, poor advice, answers that sounded completely credible but weren't. With people, we've spent decades building management structures around this. There is a manager, there are controls, approval limits, performance reviews, and escalation processes when something goes wrong.

But introduce an AI agent into that same workflow and, in many organisations, those basic management questions haven't been answered. There may be governance around data security, privacy and whether a tool can legally be used. That's important. But who is managing the quality of the work? That's a different question.

Maybe we need to start managing agents like we manage roles

I'm not suggesting HR needs to own the technology, it probably shouldn't, but HR does need to have a much stronger voice in defining the work.

If an AI agent is going to perform part of a role, we should be able to answer some fairly basic questions:

• What exactly is the agent responsible for?
• What can it decide independently?
• What requires human judgement or approval?
• Who owns its performance?
• How are errors identified?
• What happens when something goes wrong?
• At what point does the work get escalated back to a person?

None of that is particularly technical, it's essentially organisation design. We already think about spans of control, decision rights, job scope, accountability and escalation for people. Why wouldn't we apply the same thinking to agents doing work alongside them?

If HR doesn't help shape that framework, someone else will. The risk is that the conversation then becomes almost entirely about productivity and technology rather than the people affected by those decisions. There's another issue we probably aren't talking about enough: Managers themselves are going to have to change.

We're moving towards a world where managers won't simply manage teams of people. Increasingly, they'll be managing work delivered through a combination of people, technology and AI agents. That's a very different management environment and there's an interesting timing issue here.

Some of the managers we're asking to lead through this change started their careers during or around the pandemic. Many had far less exposure to the informal coaching, observation and face-to-face development that previous generations of managers experienced.

Now we're potentially asking them to manage a workforce that is becoming more complex, not less. Preparing them for that can't simply be an IT training programme. It has to be part of leadership and management development and that absolutely is an HR issue.

Where I've landed

I came away from Bangkok thinking the biggest AI question for HR isn't actually about the technology.

It's about ownership.

The tools will change. The use cases will change. Six months from now we'll probably be discussing capabilities that weren't even on the agenda that evening. But one question will remain: When work is being done by AI, who is accountable for that work?

For me, HR doesn't necessarily need to own the agent, but it should absolutely help define the role, the accountability around it and how that work interacts with the rest of the workforce.

Right now, I'm not convinced enough organisations have worked that out and as agents move from helping people do their jobs to actually doing parts of those jobs themselves, we're going to need an answer fairly quickly.

Our Bangkok roundtable report looks at this in more detail, including what HR leaders in the room told us and six operating principles for addressing the ownership gap.