Hotel Marketer Recants on AI Agents: 'They Change How We Work'

A hospitality marketing professional has publicly reversed his position on AI agents, arguing they represent a step-change beyond generative AI because they execute complete tasks rather than merely produce content. "After using Bots / Agents for a few weeks, I realize it will change how we work — completely," the columnist wrote in Tell, his weekly newsletter on hotels, tech and marketing.

His earlier skepticism centered on Openclaw, which he dismissed as a curiosity for tinkerers. "Openclaw was cool if you liked to tinker with software tools," he wrote. "But for the rest of us, it was a weird science project." Weeks of hands-on use changed that verdict.

What separates an agent from generative AI?

The distinction he draws is execution. Generative AI produces text, images, HTML and code. An agent (software that pursues a defined goal autonomously, logging into systems and acting through the screen when no API exists) completes the whole task. He frames agents as expert team members that work until the job is done, suggest alternatives when the first approach fails, and absorb the boring work nobody wants.

His central example is advertising. Today a hotel can ask AI to analyze its ads. An agent, by contrast, runs the campaign end to end: it requests the data it needs, launches the work, sends daily updates with suggested changes, and implements them upon approval. "Giving twenty small autonomous workers actual goals and letting them figure out their way, that is something else," he wrote.

Which hotel use cases did he sketch?

He offered four operational examples, each with a caveat about security and privacy. He was explicit: "None of these are great ideas in terms of security and privacy. They're here to illustrate what could be done."

• Room assignments. The agent would review nightly arrivals against stay history, preferences, past complaints, room moves, loyalty status, housekeeping constraints and maintenance blocks, then suggest assignments and implement them on approval. His framing: the PMS (property management system) stores data, but the agent reasons across it.
• Night audit. At 2 a.m. the agent would compare payments against reservations, flag suspicious balances, identify failed charges, draft manager notes and the morning report, escalating only unclear cases.
• Profile deduplication. Between 4 a.m. and morning, the agent would analyze 300 possible duplicate profiles per night, comparing names, emails, phone numbers, addresses, stays, transactions, notes and loyalty data. Confident matches merge under rules, with the profile holding the most stays or transactions as master; uncertain cases go to a review queue.
• Guest invoice requests. The agent monitors the inbox, identifies the guest, searches the PMS, retrieves or generates the invoice and drafts or sends the reply, asking the guest for booking details when the search fails.

Marketing, he noted, offers safer applications than guest-data operations.

Does this replace hotel staff?

His answer is no. "I don't think this will replace people, I think this gives everyone super-powers that will make the whole team exponentially faster and more capable," he wrote, while stressing that humans must still supervise the bots. He leaves open a harder commercial question: how much software hotels need at all if a bot can simply perform the task. He also suggests agents could troubleshoot problems mid-trip when things go wrong. For an industry chronically short of night staff and data hygiene hours, the economic appeal is obvious, even if the privacy homework remains undone.