This insight summarises I think I was wrong about Agents, by Martin Soler, Partner at Soler & Associates, published on Hospitality Net on 5 November 2026. The column first appeared in Soler's newsletter, Tell.
A change of mind. Soler had been sceptical, but now argues that agents are to AI what GPT was to the pre-AI world, and describes them as "expert team members" who are patient and willing to do tedious work. Soler adds that OpenClaw appealed mainly to people who like to tinker with software and was "a weird science project" for everyone else.
Goals replace prompts. Asking AI to analyse ads is different from telling an agent to run a campaign to increase sales. The agent logs in, requests the data it needs, sends daily updates with suggested changes and implements them once approved. If a system has no API, it can work through the screen. In Soler's framing, the PMS "stores data" while an agent "could reason across it."
Four hotel examples. For room assignment, an agent reviews arrivals each night against stay history, preferences, past complaints, loyalty status and maintenance blocks, then implements its suggestions after approval. For night audit, it checks payments against reservations at 2 a.m., flags suspicious balances, drafts the morning report and escalates only unclear cases. For profile deduplication, it works through possible duplicates, merges confident matches and queues uncertain ones. For guest invoice requests, it finds the stay, retrieves the invoice and drafts or sends a reply.
The author's caveats. Soler says none of these examples is ideal from a security or privacy standpoint; they illustrate what is possible. Marketing, Soler says, offers "much safer use cases." Soler does not expect agents to replace people and says humans must still supervise them. The column also asks how much standalone software will be needed if bots can perform tasks directly.
AIHA practical takeaway. Start with a bounded, auditable task where people approve the output and exceptions go to a review queue. Before an agent touches guest data, settle which systems it can reach, whose credentials it uses, what it may change without approval, how a merge or payment action is reversed and who owns the queue. Working through the screen can sidestep controls an API would enforce, so check access rights and vendor terms first.
Related AIHA insight: When the Hotel's AI Acts, Who Is Accountable?