AI for pool service companies: routes, chemistry logs and repair quotes.

A pool service company sells the same visit hundreds of times a season, and each one finishes the same way: the technician tests the water, decides what it needs, adds it, and writes down what they found and what went in. That record is the company's account of what it did at someone's pool. Assistants are worth adding to a pool company's office only if they keep that account complete and turn the technician's notes into quotes and messages sooner, while every decision about the water stays with the person standing beside it.

In practice the assistants take four jobs: checking each visit's log for missing readings and readings outside the ranges the owner wrote down, drafting repair quotes from the technician's diagnosis and photos, drafting customer messages, and keeping a record of what the technician observed at gates and drain covers. They do not choose a chemical, an amount or a moment to add it, decide whether anyone can swim, diagnose equipment, set a price or tell a customer that a pool meets any rule. A flag from a Bot says a reading is outside the owner's range; it never calls water unsafe or a pool compliant.

Consider, as an illustration used throughout this article, a company with an owner who still runs a route, two technicians on weekly residential routes and a part-time office manager. The owner's stop sheet, written years ago and refined since, has the technician skim, brush, empty the baskets, read the filter pressure, test the water, add what it needs and record the result, then photograph anything worth a second look. None of that is new. The opportunity is in the work after the stop, which today the office manager retypes at six in the evening.

The reason this trade's back office should keep its hands off the chemistry is set out in a CDC report. That report, published in May 2019 in CDC's Morbidity and Mortality Weekly Report and read for this article on 16 October 2026, estimates 13,508 emergency department visits for pool-chemical injuries in the United States from 2015 to 2017, and places at least fifty-six percent of them at a residence rather than a public venue. Among its prevention points is a plain one: never mix chlorine products with acid, because doing so could create toxic gases.

The same report says its pool chemical safety recommendations are generally the same for residential and public venues. A company that services backyard pools and a community pool therefore handles chemicals with the same care at both, under the owner's written handling rules. The water ranges are another matter: a public pool's come from the code its jurisdiction applies, as the companion article on chemistry logs explains. That is why no assistant in this design suggests a product, a quantity or a combination. A Bot that suggested which product to add, how much, or when would be making a handling and dosing decision from a desk, with no sight of the pool, the containers or the person about to open them.

In Grok Bot the owner names each Bot and gives it a job, lasting rules to remember and a conversation of its own, and all of them share one persistent cloud computer that provides a web browser, a file system and a terminal. A Bot uses a connector where one exists and can work many browser-based tools that have none, though a site may block automation or demand a person, and logins, two-factor codes, CAPTCHAs and payments are handed back to the user through a takeover. Cursor's help pages add a Vault, where saved logins and cards let a Bot sign in and check out without seeing them, and no setting asks for approval each time a saved card is used, so this design saves no card there. Routines run a Bot's saved workflow on a schedule, no closer together than five minutes and up to fifty per Bot. The documentation describes no phone calls, no voicemail and no text messaging, and names no connector for pool route or field-service software.

For this company the roster starts at one and grows to three. The Quote Bot comes first, because its inputs already exist and the owner reads every draft. The Log Bot follows once the owner's ranges are written down, and then a Notes Bot that drafts visit summaries and safety notices. The owner sets up each one; these three jobs come from Aevrion's pattern rather than from the product, which itself advises keeping the roster as small as the work allows. Reading visit records out of the company's route software is a custom integration or supervised browser use where the vendor permits it, subject to testing.

All three Bots use the same computer, so a supplier login saved for the Quote Bot is equally within reach of the Log Bot; separate Bots are not a security boundary. If the owner wants the parts account walled off from everything else, that takes a separate user account. And since the docs caution against treating what a Bot remembers as the record, ranges, prices and customer details come from the company's own files on every use.

Based on the Grok Bot documentation at docs.x.ai as checked on 15 October 2026. Its statements on Bot rules, Bot-to-Bot messaging, routine triggers, connected apps, approval expiry and payments were rechecked against docs.x.ai and Cursor's help pages on 26 October 2026. The trade examples are Aevrion's illustrations, not documented Grok Bot use cases. Aevrion Ops is an independent consultancy. It is not affiliated with, endorsed by or a partner of xAI, SpaceXSI or Cursor. Grok Bot is named here only to describe it.

Each evening a routine on the Log Bot reads the day's visit records and lists, for the owner, every stop where a required reading is missing and every reading that falls outside the range the owner wrote for that kind of water. The list names the stop, the reading and the range; it carries no advice. Whether a high reading needs a return visit, a word with the customer or nothing at all is the technician's call, with the owner. Because nothing in the routine sends or changes anything, there is nothing in it waiting for anyone's approval overnight.

The companion article on chemistry logs and safety observations sets out how the ranges are written, what the check flags and what it is never allowed to say, including where residential and public pools take their ranges from. Its central rule is the one this page depends on: the assistant checks the record, and the technician decides about the water.

Suppose, on a Tuesday stop, the technician finds water pooling under the pump and a heater that will not stay lit. The diagnosis is theirs: they write that the pump's shaft seal is leaking and that the heater fault is outside the work the company does, and they photograph the pump, its label and the heater's display. The photos and notes reach the Quote Bot through the owner, who forwards them into its conversation, or through a folder the owner has given the Bot, filled from the route software by a custom integration.

The Quote Bot drafts the quote in the company's format: the parts the technician named, the labor line from the owner's price list, the warranty wording copied from the owner's document, and a referral line for the heater in the owner's own words. Every price is read from the price list on each draft. Where the customer's service agreement sets an amount below which the technician may simply go ahead, the Bot notes which side of it this repair falls on; it authorizes nothing. The owner checks it, and any email reaches the owner as an editable draft to send or throw away.

Once the customer agrees, the Quote Bot can prepare the parts order, and the owner's own approval rule for purchases stops it for a yes; buying is one of the steps the docs suggest keeping behind an approval, and an approval never reverses work already done. The opening-season article shows how a one-person company runs those orders through a spring of worn parts.

Should the three Bots ever share a group chat, the photos would not travel through it: a Bot's handoffs to a group carry text only for now, so an image goes straight from one Bot to the Bot that has to inspect it. Cursor's help pages say no switch turns off messaging between Bots; a Bot follows instructions about it.

A technician on a weekly route is inside someone else's pool enclosure every week, so they notice what a homeowner may not: a gate that no longer swings shut, a door alarm switched off, a cracked drain cover. CPSC's booklet on residential pool barriers, read on 16 October 2026, cites CDC figures that drowning is the leading cause of unintentional death for children aged one to four, and says its own barrier guidelines are not mandatory unless a state or locality has adopted them. So the company's answer is an observation, never a verdict: the Notes Bot turns the technician's note and photo into a dated entry and a notice for the owner to send, and no draft calls a pool safe or compliant. The companion article on chemistry logs and safety observations sets out what is observed, the drain-cover question and how the notice reads.

Training is the owner's decision too. The Pool & Hot Tub Alliance, a nonprofit trade association for the industry, lists courses for pool operators and for service technicians; which a technician should hold, and whether a state requires any, is for the owner to establish. The business calls around each visit sit with the owner as well.

• THE EQUIPMENT

  Diagnosis is the technician's; anything outside the company's work is referred out by the owner.

• THE PRICE AND THE GO-AHEAD

  The owner sets prices, the customer approves the repair, and the owner approves each parts order.

• EVERY CUSTOMER MESSAGE

  Quotes, summaries and notices are drafts the owner sends.

• CREDITS AND REFUNDS

  A disputed visit or a refund is the owner's decision alone.

In this illustration the evening retyping is the place to look first, and the repair quote is the cleanest part of it to hand off: the technician's diagnosis, the photos and the price list already exist, and the owner reads every draft before it goes. The Log Bot follows once the ranges are written down; the chemistry-log article shows how to test them before any routine runs.

Check that starting point with the workflow automation readiness assessment. It costs nothing, asks for no email address, and scores one workflow across twelve fixed dimensions, giving a result straight away along with the obstacles it identified and a suggested next step. Score the repair-quote workflow; if the price list covers parts but not labor, or the service agreements never state the amount below which a technician may go ahead, expect rule clarity to come out low, and a low score there caps the overall result. Once those are written, a conversation with Aevrion Ops can plan the three Bots and the custom integration to the route software, which depends on the vendor's API or terms and on testing.

1. Every pool visit ends in readings and a dose; assistants check the record, and the technician decides what the water needs.
2. CDC's 2019 report estimates 13,508 emergency visits for pool-chemical injuries from 2015 to 2017, at least fifty-six percent of them at a residence.
3. No assistant suggests a product, an amount or a mix; CDC's advice is never to mix chlorine products with acid.
4. Repair quotes are drafted from the technician's diagnosis, photos and the owner's price list, and parts orders stop for a yes under the owner's own approval rule.
5. Gate and drain-cover notes become dated observations and notices; no draft calls a pool compliant or safe.
6. CPSC's barrier guidelines are not mandatory unless adopted locally; the technician observes, the owner notifies and the local authority decides what applies.