Booking autumn in July: a two-sweep chimney company's quiet-season office.

For a chimney company the slow season is the only time the autumn can be arranged. Households whose fireplaces and stoves will be lit in October can be offered August and September while the diary is empty, and the offer reads better when it speaks to the appliance they actually own. This article follows one illustrative July: the customer list split by appliance type, a reminder draft for each group that the owner reviews and sends, and the quiet weeks used to settle the report format before the rush.

Everything about this company is invented to make the design concrete: its vans, its customers, its July. It reports no outcome and stands for no real firm.

In this illustration the owner and one employed sweep each run a van. The customer list is a spreadsheet exported from last season's booking software, the employed sweep keeps visit notes on a phone, and the owner writes reports and reminders in the evenings. In October there are no evenings left.

The arrangement comes from the One-Man Army pillar, in which a single decision-maker hands each recurring desk job to its own named AI assistant. Here the owner creates three Bots in Grok Bot under Aevrion's pattern: a Recall Bot, a Diary Bot that drafts the autumn booking grid, and a Report Bot.

All three use the one persistent cloud computer that every Bot on the owner's account shares, so the spreadsheet the owner places in the shared workspace is readable by each of them. The same sharing means separate Bots are not a security boundary, and the owner decides what goes onto that computer. Standing rules are given to each Bot in chat to remember, which guides behavior without enforcing it.

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.

The first job of July belongs to the Recall Bot: reading the spreadsheet and grouping every customer by appliance and fuel. It reads the spreadsheet afresh each time rather than trusting last year's impressions; the documentation is clear that facts which change belong in the source system, not in memory. Where the record says nothing, the customer goes into an unknown group for the owner instead of being guessed at.

Each draft, whatever its group, closes with the intake question the money guide for this trade recommends, asked in July instead of October: has anything about the fireplace or stove changed since the last visit? A new appliance or a change of fuel is among the triggers a published restatement of NFPA 211 lists for a Level II (the standard itself is sold by NFPA and was not read for this guide), the longer visit that may include a camera inspection of the flue. An answer received in summer lets the owner book that visit at its proper length while August still has room.

The reason given to the gas group comes from CSIA's homeowner guidance (as checked on 15 October 2026). The draft states it once, and the owner can cut it.

The Recall Bot drafts the email for each group in the owner's words, filling the household's name, appliance and last visit date from the record. Every one appears in the Recall Bot's own conversation as an editable draft that the owner reviews and then sends or discards. The documentation describes no telephone or text-message capability, so households who never read email stay on a short list the owner works by hand.

In this illustration the owner releases one group at a time: the gas group first, because those visits are short and pack into mornings, then the wood stoves, whose longer visits need the most planning. Staggering the sends staggers the replies, and one person reads every reply.

A routine on the Recall Bot runs once each weekday morning and rebuilds three lists: who has replied, who has booked, and who has heard nothing for ten days. Routine schedules have to be at least five minutes apart, so a daily run is nowhere near the limit. The routine's instructions say that if the spreadsheet cannot be opened it should report that, and never fall back on a stale copy.

As replies come in, the owner passes each preferred week to the Diary Bot. It drafts the autumn grid for both vans: visit lengths by appliance as the owner set them, households grouped by area so a van is not crossing town between a stove and a gas flue, and two half-days a week left unbooked for work that cannot wait for November, such as a house sale or a flue after a chimney fire. The grid is a proposal. The owner moves whatever needs moving and approves it in the Diary Bot's conversation.

The employed sweep's days are discussed with the owner, face to face. Because the three Bots sit on the owner's account, the second sweep has no way to approve or send from them; their knowledge of their own customers reaches the grid through the owner.

A group chat holds all three Bots (a group can contain between two and six). Taking its cue from a documented example in which a group's coordinator can hand out work, the Diary Bot plays that part: it posts each draft week and asks the Recall Bot which households have confirmed. Outside the group, one Bot may write to another or pass it a job, and Cursor's help pages say no switch turns that messaging off.

Groups never show approval cards or email drafts, so every send and every approved grid happens in the owning Bot's own conversation. Handoffs that Bots post to a group are text-only for now, so when the Diary Bot needs to see a photo of a stove that the Report Bot has on file, the Report Bot sends it Bot to Bot directly rather than through the group. The Chief of Staff in the documentation's examples writes a digest for a person; nothing in the documentation describes a Bot supervising other Bots. Making the Diary Bot coordinator is Aevrion's use of documented pieces, and every week it proposes is still read by the owner.

The second half of July is when the owner fixes the report. The owner attaches a handful of last season's reports from each sweep in the Report Bot's conversation and asks for a comparison: where the two describe the same component differently, which fields one fills and the other skips, and where their photos fall in a different order. The Bot returns a side-by-side list, each difference tied to the report it came from, and proposes one field order.

The two sweeps settle the wording together, with the owner, because the words in a finding are the sweep's professional statement and not the office's. The agreed format is saved as a skill the Report Bot uses every time, stating when to use it, what it needs, the field order, how to check a draft and the rule that it never writes whether a chimney is safe to use. In October, every draft starts from that format, and the sweep who made the visit finishes it.

At the end of August the Diary Bot lists October's empty slots by van and the Recall Bot lists who has not answered, by group. The owner decides the second wave: a shorter note to the silent wood-stove households, a personal message to customers of many years, and nothing to anyone who has asked not to be contacted.

The sweeps decide what September is for. Some held half-days can be released to the waiting list; others stay held, because a chimney fire in the first cold week will not book itself two weeks ahead. When the first cold evening of October brings its inquiries, the owner can answer them from a grid that already exists instead of building one by hand at night.

Open last season's customer list and try to sort it by appliance and fuel. If a large part of it cannot be sorted, that gap is the first job, and this summer's visits and old reports can fill it.

Next, score the summer recall, and only that, in the workflow automation readiness assessment; it asks for no registration and rates the one workflow on twelve dimensions. Expect data availability to score low if appliance and fuel were never recorded per customer; the result will say so, name the other blockers and suggest a next step. Once the list is sorted, Aevrion Ops can work through the groups, the three Bots and their approval points with you, plus a custom integration to your booking software, built only if that vendor's API or terms allow it and tested before use.

1. A two-sweep company cannot add vans for October, so it moves October into the summer, one appliance group at a time.
2. The recall list is grouped from the customer record, not from a Bot's memory; unknown households go to the owner.
3. Every July email asks what has changed, because a new appliance or fuel can turn a routine visit into a Level II.
4. Reminder drafts are sent, and the autumn grid approved, by the owner in each Bot's own conversation, never in the group.
5. The quiet weeks settle one report format, agreed by both sweeps and saved as a skill the Report Bot follows.