Four organisations weighing AI agents for HR say the hard part is often not the software, but fixing processes, preparing data and getting legal approval. Their experiences offer a useful reality check for businesses hoping to add agents to recruitment and other sensitive workplace decisions.
Watch Desk analysis
What happened
At Oracle AI World, IHG Hotels & Resorts, UL Solutions, Deutsche Bahn and Acosta Group described different stages of their HR technology plans. Their comments were reported by Diginomica, which disclosed that Oracle was a partner at the time of publication.
IHG is considering talent acquisition as a starting point for agentic AI, but its HR technology director said the exercise has also exposed process problems, including gaps in interview scheduling. UL Solutions is reviewing earlier implementation choices and working on the data needed to support AI and other initiatives. Deutsche Bahn’s HR IT chief advised starting data cleansing early, while Acosta said it brought legal into discussions about AI language in job postings.
The customers also described practical costs beyond the initial setup. UL Solutions’ representative said agent deployments can require substantial testing and debugging, and that token costs can be complex. At IHG, legal review depends on the use case: interview scheduling was treated differently from a tool matching candidates to jobs, where questions about the system’s purpose and human involvement become more pressing.
Why it matters
These examples make clear that an AI agent does not arrive in an empty office. It inherits existing processes, data and approval structures, including the awkward bits organisations may have hoped to automate away. Poorly prepared systems can turn a familiar workplace problem into a faster-moving one, not necessarily a better one.
The decisions also vary with what an agent is allowed to do. Scheduling an interview is not the same as ranking candidates. Deutsche Bahn’s account adds a useful counterpoint to the idea that regulation is only a brake: its representative said the work required by EU rules can force the organisation to think through its plans and build a sound technical base.
Our read
The most useful lesson is refreshingly unglamorous: sort out the process and the data before handing the job to an agent. That is not a reason to abandon the technology. It is a reminder that “we have an AI tool” is not the same thing as “we have a workable HR process”.
The examples are customer accounts, not proof that agentic HR is already producing better hiring or lower costs. But they give buyers specific questions to ask: what decisions will the agent make, what needs fixing first, how will its work be tested, and who gets a say when employment decisions are involved?
What to watch
- Whether IHG proceeds with talent-acquisition agents, and which tasks it gives them.
- How customers prepare data and measure the testing and token costs of deploying agents.
- Whether legal teams and works councils are involved early enough to shape the use cases.
Discussion spark: Should employment-related AI face formal legal review before deployment, or should the level of scrutiny depend on the specific task an agent performs?
Sources and evidence
- Want to hear more about AI adoption? (28 October 2026, 15:28 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.