On May 5, Anthropic released ten Claude agent templates for financial services. The list reads like a job description for a junior analyst at a bank. Pitch builder, meeting preparer, earnings reviewer, model builder, market researcher, valuation reviewer, general ledger reconciler, month-end closer, statement auditor, KYC screener.
This is the playbook shift. The frontier labs have stopped selling models and started shipping vertical agent libraries. Hospitality is next. The ten agents that should exist for hotels are not the same ten that show up in most slide decks.
Why this format is the format
Until last week, most enterprise AI vendors sold a model and a developer kit. Build your own agents on top of us. The pitch was that infrastructure was the product.
The shift Anthropic just made, and that the rest of the field will copy within the quarter, is to ship the agents themselves with the model under the hood. The buyer is no longer a developer team. The buyer is a business owner who can recognize "month-end closer" as something they already pay a person to do.
That changes what hospitality vendors should be building. Not a Claude wrapper. Not a custom GPT. A library of named agents that an operator can recognize, evaluate, and assign work to.
Canary Technologies started this work in March with their Hospitality AI Agent Studio launch, shipping pre-built Front Desk, Concierge, and Central Reservations agents. That is the right shape. The list below extends the idea to the ten roles a hotel actually has to fill in 2026. (For the discovery-layer counterpart to this on-property agent list, see Boutique hotels and the coming AI discoverability layer.)
The ten agents that should exist for hospitality
These are not the agents that show up in most product roadmaps, which tend to fixate on the booking surface and the chatbot on the homepage. These are the agents that map onto the work hotels actually do.
The pre-arrival agent. Reads the reservation, the loyalty profile, and any prior stays, and prepares a personal arrival plan before the guest lands. Knows the guest's room preference, dietary notes, transportation choice, and which restaurant they liked last time. The work this replaces is the one a great front desk supervisor used to do quietly the morning of arrival. Most properties stopped doing it when the supervisor left.
The on-property concierge agent. Answers any question a guest can think of during the stay, in their language, instantly, with full context of where they are and what is open. The bar here is that the agent has actually read the property's information, not the generic one a hotel chatbot pulls from the website footer.
The rate-and-inventory agent. Talks to other agents on the demand side. When a guest's agent asks for availability and a fair rate, this one answers in real time, holds the room, and negotiates edge cases (early check-in, room type substitution, group bookings) without a human in the loop. This is the inbound side of the agentic booking shift, and the property without one will lose bookings to properties that have one.
The review-and-recovery agent. Reads every review on every platform within minutes of posting, drafts a response in the property's voice, flags the ones that need a real human to respond, and routes the operational issues (broken AC in 412, slow check-in on Friday) to the right department. Most hotels are still doing this on a 48-hour delay through a marketing manager.
The housekeeping coordinator agent. Reads the day's check-ins, check-outs, room assignments, and special requests, and produces a sequenced housekeeping plan that gets rooms ready in the right order for arrival times. Talks to the front desk agent when plans change.
The maintenance triage agent. Receives every reported issue (guest complaint, housekeeper note, sensor reading), categorizes it, decides what is urgent, and creates a work order for the right technician. Answers the guest with an honest ETA. This is the agent every hotel claims to have and almost none actually do.
The sales-and-events agent. Reads inbound corporate inquiries, qualifies them against the property's calendar, and drafts a real proposal with rates, room blocks, and event-space options. The sales coordinator hates losing the lead because the email sat in their inbox over the weekend. This agent does not let that happen.
The forecasting and pricing agent. Pulls demand signals (events in the city, competitor pricing, historical patterns, weather), and proposes rate moves with reasoning the revenue manager can actually evaluate. Not a black-box yield engine, an agent that explains its logic.
The lost-and-found and post-stay agent. Closes the loop on every guest after they leave. Confirms the receipt is correct, asks for the review, handles the lost-item return, surfaces a personalized reason to come back.
The compliance-and-policy agent. Reads every guest interaction, every transaction, and every policy change, and surfaces issues before they become incidents. The hotel-industry equivalent of the KYC screener Anthropic shipped for finance.
What is true of every agent on this list
Three things, and they are the part nobody pitching a hotel-AI product in 2026 should get to skip.
Each agent has a job a real human used to do. The list above is not abstract capabilities. It is a payroll. The reason the financial-services list works is that every agent on Anthropic's list maps onto a junior banker's daily task list. The hospitality version has to do the same, or it does not ship.
Each agent has access to the property's actual systems. Not a generic LLM with a system prompt about hotels. The pre-arrival agent has to read the PMS. The maintenance triage agent has to write to the work-order system. The rate agent has to talk to the channel manager. The hotels that get value from agentic AI are the ones whose data layer is honest, current, and connected. The rest will get demos, not outcomes.
Each agent has a brand voice. A guest does not want to talk to "the hotel's AI." They want to talk to the hotel. The pre-arrival agent at one property should not sound like the pre-arrival agent at another. This is the part most platforms get wrong out of the gate, because they ship the agent and ask the operator to add the voice later. The voice is the agent. It is not a skin.
What this changes for the next 12 months
The vertical-agent-template era started for finance on May 5. It will start for hospitality before the end of 2026, either from Anthropic, OpenAI, or one of the platform companies already serving the industry.
When it does, the question for every hotel is not "should we adopt agents." It is "which library do we adopt, and which of these agents do we want building muscle around our brand voice and our data?" The hotels that wait until the library shows up to figure out the answer will be a year behind the ones who started thinking about it now.
If you run a property and you are starting to map your operation onto something like the list above, we would love to compare notes. The list is not the answer. It is the conversation the industry should be having, and almost nobody is.
The finance starter pack shipped this week. Hospitality is next.