A six-week spring backlog with one operator: keeping fence jobs moving.

A fence crew can only install what has been cleared, so a spring backlog is really two queues wearing one name. One is visible: jobs waiting for crew days, laid out on the calendar. The other is not: jobs waiting on an association committee, a permit office, a utility locate or a boundary question, which surface only when the crew turns up and cannot start. Chasing that second queue is desk work, and for a company with one operator it is the desk work that gets dropped first.

Supervised AI assistants can keep the second queue in plain view, draft every chaser and check each week's installs against the approvals they need. The owner sends, signs and decides. This article follows one invented company through one spring to show the design; no real firm stands behind it, and it claims no result.

The illustrative company is based around Orlando, Florida, and its owner quotes in the evenings, measures on Saturday mornings and leads a two-person install crew on weekdays; nobody else works in the office. By the end of March the signed jobs fill six weeks of crew days, and each one carries its own outside approvals. A privacy fence inside a homeowners' association needs the association's yes. A pool enclosure needs a city permit, because the city's guide, read on 16 October 2026, lists a fence used as a mandatory pool or spa barrier among its permit triggers. Every job needs utility marks before a post hole.

Behind this design sits the model described in the One-Man Army pillar: the owner keeps every decision, and every repeating block of office work is handed to its own named assistant. Here the owner builds three Bots in Grok Bot, following Aevrion's pattern. Approvals owns the chase. Crew Week proposes the install order. Supply tracks material orders and deliveries. He starts with Approvals alone and adds the other two only once its tracker matches what is really happening.

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.

Take one job from the illustration. A picket fence inside an association is pencilled in for the third week. At signing, the homeowner said they would send the application to the architectural committee. On the Friday before the install, the owner rings to confirm and learns it was never sent. He reaches for the next job in line to fill the slot, and finds nobody has requested its locate. Two crew days are lost, and neither loss had anything to do with the crew.

So the Approvals Bot's first job is a re-sort. It reads every signed row in the company's shared tracker and groups the backlog by who each job is waiting on, not by contract date: association committees, the permit office, locates not yet requested, boundary questions still open. Jobs with no recorded chase date go to the top. The Bot reads the tracker afresh each time; the documentation tells users to keep changing facts in their own system rather than in a Bot's memory, so the spreadsheet stays the record.

Association applications can be the slowest to chase and the easiest to forget, because they are not the company's to submit. They go in the homeowner's name and are signed by the homeowner, and that stays human. What the Approvals Bot can do is assemble whatever packet the association's own guidelines ask for, in this illustration a site plan with the fence drawn on the survey plus its style, height and color, and draft a short covering note and a reminder to the homeowner. Committee dates come only from the association's own published documents.

Each draft appears in the Approvals Bot's own conversation as an editable email, and the owner chooses Send or Discard; a crew member can be asked about a job but cannot send from the owner's Bots. Permit portals and supplier sites are supervised browser use where the site allows it, subject to testing: the documentation says a site may block automation, and for any login, two-factor code or fee payment the owner takes over the Bot's screen himself. 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.

Once the tracker is trusted, the owner asks the Approvals Bot for a routine: every Thursday evening, read the tracker and test each of next week's installs against five gates. Thursday is the illustration's choice, because it leaves Friday to fix what fails. The routine is documented product behavior, scheduled by asking the Bot and confirming its details; the gates are Aevrion's pattern. Its output is a list and a set of drafts, never a decision.

Its limits shape how the owner uses it. Schedules cannot run closer together than five minutes, which matters not at all for a weekly check. A routine starts without the owner, so if one of its runs raises an approval request while he is setting posts on a job, the request expires after roughly ten minutes (Cursor's help pages say thirty) and that step does not run; requests he meets in his own chats wait for him. That is why this routine is written to read and draft only. The docs mention no retry and no alert when a run fails, so on Friday morning the owner opens the routine's run records before trusting an empty list.

A job passes the gate only when all five are true:

• ASSOCIATION

  Written approval is on file, or the owner has recorded that no association applies.

• PERMIT

  The permit number is on the row, or the owner has recorded that no trigger applies.

• LOCATE

  A ticket is logged and every utility on it has responded.

• LINE

  The basis for the fence line is written down: a surveyor's stake-out, or a recorded survey whose markers were checked against it on site.

• MATERIALS

  Delivery is confirmed for a day before the install.

Friday morning, the list shows one failure: a vinyl fence whose locate ticket still has a utility outstanding. Crew Week proposes the earliest job that passes all five gates, and drafts two messages, one telling the first customer plainly what the job is waiting on and one offering the second customer an earlier start. The owner reads both and sends them. Whether the ground is fit to dig after a week of rain is his call on the morning, not the routine's.

A group chat holding all three Bots works as the backlog board. The documentation allows two to six Bots in a group and gives the example of a coordinator inside one that can assign work, so Crew Week posts the week and hands out tasks. Treating it as the board's coordinator is Aevrion's use of those documented pieces; a supervising chief-of-staff Bot is not a product feature, and the docs' Chief of Staff example is a digest written for a person. Bots can also message each other and pass a job between them, visibly, and Cursor's help pages say no switch turns that messaging off.

The board can only flag approval cards and email drafts, since neither ever shows up inside a group; the owner sends from each Bot's own conversation. A handoff a Bot posts to the group can currently carry text only, so a photo of a delivered pallet goes straight from one Bot to another. All three Bots share one cloud computer and its sign-ins; separate Bots are not a security boundary, so a supplier sign-in stored on it is usable by any of the three.

Orlando's guide keeps every fence on or within the lot's own boundary. In this illustration the owner relies on a recorded survey whose corner markers he has found and matched to it, or on a licensed surveyor's stake-out, which the homeowner engages whenever the survey and the ground disagree. The Approvals Bot files the survey and marks the line gate as passed only when the owner records the basis; it never reads a boundary from an aerial photo, however tempting that is in week five.

The ground-break go is the second gate that never moves. Each install morning the owner, consulting his crew lead, decides whether the crew digs, once every utility has responded and the line is settled. The routine's Thursday list informs that decision; it does not make it.

The quiet weeks of early summer are when the tracker proves useful a second time. Each closed row holds dates the company recorded itself: when each association application went in and when the answer came back, which permit applications needed resubmitting, which jobs waited longest for a stake-out. These are the company's own observations, not rules, and the Approvals Bot can summarize them by association and by city into a short note the owner keeps.

Next March, that note changes when the chase starts. An association that took weeks to answer gets its packet drafted the day the contract is signed, and a city whose portal kept rejecting site plans gets the owner's own checklist attached before he applies. The backlog will still fill six weeks; the jobs inside it will have started waiting earlier, with someone chasing.

Before the next rush, run the Thursday gate by hand for three weeks: each Thursday, test next week's installs against the five gates on paper, and note which gate fails most and how late the failure was found. That paper routine is the one the Approvals Bot would run, and it shows whether the tracker already holds the answers.

Aevrion's readiness assessment for workflow automation is a sensible next step: free, ungated, and focused on a single workflow, scored against twelve fixed dimensions. Score the Thursday gate check, and expect data availability to drag if locate responses and association letters sit in the owner's texts and inbox rather than on the row, and error sensitivity and controls to need care, because a gate passed in error sends a crew out to dig. Its answer is immediate: a score, a result band, the blockers and a next move. From there, Aevrion Ops can design the three Bots, the gate routine and their approval stops with the owner, and build any custom integration to the tracker or supplier systems only where the vendor allows it and after testing.

1. A spring backlog is two queues: crew days on the calendar, and approvals nobody can see until the crew arrives.
2. Association applications go in the homeowner's name and stay human; a Bot assembles the packet and drafts the reminders the owner sends.
3. A weekly pre-dig gate tests next week's installs for association approval, permit, a fully answered locate, a recorded line basis and confirmed materials.
4. The routine only reads and drafts, because an approval request raised by a routine expires after about ten minutes (Cursor's help pages say thirty), and there is no documented retry or failure alert.
5. The ground-break go stays with the owner, and the line is taken from a licensed surveyor's stake-out or a recorded survey checked on the ground.
6. The season's own approval dates, recorded in the tracker, decide when next spring's chasing starts.