AI
AI Rebooking Agents for Burnaby's Atmospheric River Week
When a strong atmospheric river settles over the South Coast, outdoor work in Burnaby doesn't slow down gradually. It stops all at once. Landscapers, roofers, window cleaners, painters, concrete crews and gutter installers all face the same problem: a full week of booked jobs, a phone that won't stop ringing, and a crew waiting to hear where they're going tomorrow.
The rain isn't the expensive part. The scramble is. Someone has to call 40 customers, find new slots that don't overlap, update the calendar, tell the crew, and then do it all again when Thursday's forecast changes. That's repetitive, rules based work, and AI now handles it reasonably well as long as you give it clear limits.
This post covers what an AI rebooking agent does, what it costs to run, and when it isn't worth setting up.
What an AI rebooking agent actually does
Without the buzzwords, it's a language model connected to a few of your business tools. Anthropic's documentation calls this tool use. The model decides when to call a function you've defined and returns a structured request. Then your own software runs it. The docs say client tools 'run in your application', and your code sends the result back to the model (Anthropic, Tool use with Claude).
That matters for a trades business. The AI never edits your calendar directly. It asks, your system checks the request, and only then does anything change.
A typical rain week workflow looks like this:
- Pull the active weather alerts for your service area every few hours.
- Flag booked outdoor jobs that fall inside the alert window, using rules you set. For example: no roofing during any rainfall warning, and exterior painting paused if rain is forecast within 24 hours.
- Text or email each affected customer two or three open slots taken from your real calendar.
- Read the reply, book the slot they pick, and confirm it in plain language.
- Pass anything unusual to a person: upset customers, deposits, warranty questions, and jobs tied to a permit or inspection date.
- Send the crew lead an updated run sheet each evening.
None of this is magic. It's the same routine your office manager already follows, except it can run at 2 a.m. without overtime.
Use official weather data, not guesses
The weakest version of this agent guesses at the weather. A better version reads official alerts. Environment and Climate Change Canada publishes a Weather Alerts collection through its MSC GeoMet API. It covers warnings, watches, advisories and statements, and the data comes in formats including GeoJSON and JSON (ECCC, MSC GeoMet Weather Alerts). The collection page says warnings usually arrive six to 24 hours ahead, but fast forming storms can give much less notice.
Six to 24 hours is useful lead time for rebooking, but it isn't a week. So a sound design has two layers. The agent first proposes tentative moves based on the forecast. Once an actual alert is issued, it confirms those moves or releases the slots. Your written rules decide which alert types trigger a move. Don't let the model make up that threshold on the fly.
What it costs to run
Model prices change often, so here are the vendor's current figures. As of November 2026, Anthropic's models overview lists Claude Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens. It lists Claude Haiku 5.5 starting at $0.10 per million input tokens and $0.50 per million output tokens (Anthropic, Models overview). The tool use page adds that turning on tools adds a 286 token system prompt for Sonnet 5.5 with automatic tool choice. Your tool definitions and results add more tokens on top of that (Anthropic, Tool use with Claude).
An illustrative example, not a measured result: say one rebooking conversation uses about 20,000 input tokens and 2,000 output tokens on Sonnet 5.5. That total includes tool definitions, calendar lookups and a few text replies. At the listed prices, that's about 4 cents of input and 2 cents of output, or roughly 6 cents per customer. Rebooking 100 customers in a storm week would cost around $6 in model fees.
The model is the cheap part. The bigger costs are SMS fees, API access for your scheduling software, setup time, and a person reviewing the unusual cases. If you only rebook a handful of jobs, those fixed costs outweigh everything else.
Keeping people in charge
In December 2023, Canada's privacy regulators jointly published principles for generative AI. The signatories include the federal Privacy Commissioner and the Office of the Information and Privacy Commissioner for British Columbia (OPC, Principles for responsible, trustworthy and privacy-protective generative AI technologies). Three of the principles apply directly to a rebooking agent.
Necessity. The principles say the tool 'should be more than simply potentially useful.' If you rebook five jobs a season, you probably don't need an AI agent.
Openness. They ask organizations to 'clearly communicate to any affected party whether a generative AI tool will be used as part of a decision-making process.' Rescheduling is low stakes, but you should still tell customers they're texting with an automated assistant and make it easy to reach a person.
Limiting collection. Collect 'only what is needed.' The agent needs a name, a phone number, the job address and the booking. It doesn't need payment details, door codes or notes about someone's health. Keep that information out of what the model sees. Most BC private sector businesses fall under the province's PIPA rather than the federal privacy law, so check OIPC BC's guidance too.
For broader risk management, NIST's voluntary Generative AI Profile (July 2024) is a companion to its AI Risk Management Framework. It's a useful reference even for a small shop. In practice, that means logging every message the agent sends, testing it on fake customers before launch, and reviewing a sample of real conversations each week.
Where this doesn't apply
Be honest about whether this fits your business.
Low volume. If one person can rebook everyone in an hour, a shared calendar and a text template will do the job. The setup cost won't pay for itself.
Messy scheduling data. If jobs live in a notebook, on a whiteboard and in three people's heads, the AI has nothing reliable to book against. Fix the calendar first.
Site judgment. Whether a slope is safe after three days of rain, or whether an excavation can go ahead, is a decision for someone on site. The agent shouldn't make that call, and your rules should stop it from trying.
Customers who prefer a call. Some long time residential clients want to hear from someone they know. Flag them and leave them out of the automation.
Gaps in the tools. Anthropic's own docs say that when required details are missing, some models might infer a reasonable value instead of asking. For a booking, a guessed date is a wrong date. Their strict tool use option guarantees a request matches your schema. It doesn't guarantee the date is the right one. So require a customer confirmation before anything is written to the calendar.
Alerts trail the weather. ECCC notes that fast storms can bring very short warning times. The agent can't save a job that's already underway in a downpour.
A sensible way to start
Start with one service line, one storm, and drafts only. Let the agent prepare rebooking messages and proposed slots, and have a person approve them in batches. After a couple of storms, compare its suggestions with what your office would have done. If they line up most of the time, let it send routine confirmations on its own and keep people handling the exceptions.
That approach works whether you're in Burnaby, New Westminster or anywhere else in Metro Vancouver that spends November under heavy rain.
Sources
- Anthropic. Tool use with Claude. Claude Platform documentation, accessed November 2026. https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview
- Anthropic. Models overview. Claude Platform documentation, accessed November 2026. https://platform.claude.com/docs/en/about-claude/models/overview
- Environment and Climate Change Canada, Meteorological Service of Canada. Weather Alerts collection, MSC GeoMet OGC API. Accessed November 2026. https://api.weather.gc.ca/collections/weather-alerts
- Office of the Privacy Commissioner of Canada, with provincial and territorial privacy regulators including OIPC BC. Principles for responsible, trustworthy and privacy-protective generative AI technologies. 2023. https://www.priv.gc.ca/en/privacy-topics/technology/artificial-intelligence/gd_principles_ai/
- National Institute of Standards and Technology. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1). 2024. https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence
If rain weeks keep wrecking your schedule, book a free call with Autana Solutions. We'll look at how you book jobs today and tell you plainly whether an AI rebooking agent is worth it for your business yet.
Want an AI employee for your business?
We install a 24/7 AI worker for businesses in Vancouver, Burnaby, and beyond. Book a free Discovery Call.
Book a call →

