AI for irrigation companies: start-ups, repairs and two seasonal waves.

Irrigation demand does not spread itself across the year. Spring start-ups pile into the weeks after the last cold snap, winterizations pile into the weeks before the first hard one, and both waves land on a company whose technicians are already on the road. The work in between is no quieter: a zone that will not shut off, a controller that lost its program, a new install that needs a permit and a sealed plan, an assembly due for its annual test.

Supervised AI assistants can carry a good share of the paperwork around those waves. They can turn last season's notes and the current customer list into a start-up sheet for each property, turn a technician's zone notes into a repair quote priced from the owner's own list, keep a note of each service area's watering rules with its source, and lay out the remaining winterizations when a freeze is forecast. They do not design systems, install anything, record a backflow result, decide what a controller should run or choose who gets winterized first. Those decisions belong to the licensed irrigator, the licensed tester, the owner and the customer.

The first job for any back office is to sort what arrives. In an irrigation company that means five kinds of work, each with its own timing. Writing the split down, with the owner's own examples, is what lets an assistant apply it consistently. The table below is Aevrion's way of framing it for the office, not a rule from any regulator.

A start-up is short and repetitive: water back on, each zone run, heads and valves checked, the controller set. What makes the spring wave hard is not the visit; it is the hundreds of customers who all want it in the same two weeks. The useful preparation is knowing, before the technician arrives, what each system looked like when it was shut down last autumn.

Under Aevrion's design the owner creates a Start-Up Bot for this. In Grok Bot each Bot is a named, durable teammate with one job, and it works on a persistent cloud computer that holds files and browser sessions between tasks. The evening before each area's day, the Start-Up Bot reads the current customer list and lays out one sheet per property: the controller and the program the customer last agreed, the zones and heads noted at last autumn's blowout, the backflow assembly's due date from the owner's own records, and any request the customer has made since. The technician works from the sheet. Booking the area days themselves follows the draft-and-send pattern the blowout-season article sets out for autumn, and no booking email leaves until the owner sends it.

The list is read fresh every time. The documentation says plainly that a Bot's memory does not replace the source system, so a customer who canceled over the winter or moved house drops out because the record says so, not because the Bot remembers. Reading a field-service or scheduling system is a custom integration, or supervised browser use where the vendor permits it, subject to testing.

Grok Bot does not make or take phone calls and does not send texts, so a customer who rings to book still reaches the owner or whoever answers. What the Start-Up Bot can do is take the note of that call, once someone has added it to the conversation, and add it to that customer's sheet.

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.

Repairs arrive as symptoms. EPA's WaterSense Sprinkler Spruce-Up guidance, written for homeowners, describes the kinds of thing a walk through each zone turns up: heads that do not pop up fully or sit tilted, leaks at the joints between sprinklers and pipes, which can point to excess pressure, water pooling between cycles that may mean an underground leak, and valves that do not close completely and let a zone seep (as checked on 16 October 2026). In Aevrion's pattern, the technician's run-test notes are written the same way, zone by zone.

A Quote Bot, created by the owner under Aevrion's pattern, turns those notes into a draft quote: each zone's findings in the technician's words, each proposed repair matched to a line in the owner's price list, and a list of anything the price list does not cover. It reads the price list every time, and when the notes describe a repair with no matching line, the draft says so and stops; it does not estimate. Photos the technician took reach the Bot through the owner, who can attach up to six files at a time in the desktop conversation.

What the repair is, is the technician's diagnosis. What it costs is the owner's price. Whether to go ahead is the customer's decision. The quote reaches the customer only when the owner sends it.

Programming a controller is partly a local rules question. The San Antonio Water System shows how much those rules can move. Its year-round rules allow landscape watering with an irrigation system on any day, from midnight to ten in the morning or from nine in the evening to midnight. Its Stage 3 rules, which its homepage showed as in effect when checked on 16 October 2026, allow irrigation-system watering only once a week, on a day set by the customer's address, from five to ten in the morning and from nine in the evening to midnight. A schedule that was lawful in spring can be out of date by late summer.

Other utilities set their own rules, and a company working across several suburbs can cross more than one utility's boundary in a day. Aevrion's pattern is a watering-rules note per service area: the utility, the current stage, the permitted days and hours, the link to the utility's own page and the date somebody last checked it. A routine can open those pages on a schedule and flag any that appear to have changed since the last check, so the owner updates the note; a page that cannot be read is reported as unread, never filled in from last month's copy. Some sites block automated browsing, in which case the check falls back to a person.

The note informs; it does not decide. The technician and the owner set each controller with the customer, against the rules the owner has confirmed with the utility.

Autumn reverses the spring problem. The deadline is no longer the customer's convenience but the weather, and a forecast can pull it forward by days. Whoever is still waiting when the cold arrives carries the risk, so the order of the remaining list matters more than its length.

An assistant can make that order visible: which areas still have customers waiting, who asked to be early, who has been rescheduled once already. The order itself is the owner's call, made with the owner's own knowledge of each property. The blowout-season article in this series follows one licensed irrigator through an illustrated autumn of that work, including the forecast routine and the billing that follows.

In Texas a new sprinkler system is licensed, permitted and inspected work: the City of College Station, checked on 16 October 2026, limits installation to a TCEQ-licensed irrigator and a few named exceptions, and asks for a permit and an irrigation plan bearing the irrigator's seal before work starts, with an inspection at the end. Local requirements vary by city. The licensed-install article in this series takes that paperwork step by step, with the Texas rule sections.

Backflow testing is a separate cycle. Annual testing on the water supplier's calendar, with its notices and tester lists, is covered in the series' backflow testing article for plumbing companies, and an irrigation company whose customers appear on the same notices can use the same approach. Only a licensed tester records a result. Arrival days, warranty answers and any promise about a system's condition come from a person too.

Keep the roster small, as the documentation advises: a single Bot responsible for a whole outcome first, and a second only once a specialist role has settled. In an irrigation company the natural first Bot is the Start-Up Bot, because both waves start with last season's list. The Quote Bot follows once repair notes arrive in a consistent shape, and a Paperwork Bot for licensed installs comes last, because installs are fewer and every step in them is someone else's signature.

The Start-Up, Quote and Paperwork Bots work on one cloud computer with one set of browser sign-ins, and the documentation says in terms that separate Bots are not a security boundary, so the company's bank and the utility's billing portal stay off it. A routine that runs without the owner, such as the watering-rules check, is designed to finish with a note, because the documentation treats its approvals differently: when the owner is present in the chat, an approval request simply waits for an answer, whereas a request that comes from a routine, a trigger or a message from another Bot runs out after about ten minutes (Cursor's help pages say thirty) and its action is never taken. Routines are scheduled at least five minutes apart, the documentation describes no automatic retry, and each routine keeps its twenty most recent run records, so somebody reads the run history once a week in the busy seasons.

Pick the spring start-up sheet and nothing else, fill it in by hand for one area's day, then put it through the workflow automation readiness assessment. The assessment costs nothing and sits behind no form; it rates that single workflow on twelve set measures, among them repeatability, volume, data availability, integration availability and ownership clarity, and its result comes back on the spot with the blockers it detected and a suggested next move. If last season's customers are spread across a scheduling tool, a notebook and the technicians' phones, expect data availability to come out low; if nobody has written down what a start-up visit checks, rule clarity will be low too, and that caps the result.

Where the score says the start-up list is ready, a scoping conversation with Aevrion Ops is the next step: the Bots worth creating first, where each one stops for the owner, and whether a custom integration to the company's scheduling software is possible under that vendor's API or terms, to be built only after testing.

1. Irrigation work arrives in two compressed waves, spring start-ups and pre-freeze winterizations, with repairs, installs and backflow cycles running between them.
2. A Start-Up Bot can lay out a sheet per property from last season's notes and the current customer list before each area's day; the technician works from it.
3. Repair quotes are drafted from the technician's zone notes and the owner's price list; anything without a price line goes back to the owner.
4. Watering rules differ by utility and drought stage, so each service area needs a dated note linked to the utility's own page.
5. Freeze order, design, installation, backflow results and controller schedules stay with the owner, the licensed irrigator and the licensed tester.