Skip to content
be-human.ai
← Back to the journal

AI by industry

Superintelligence Prep for Window Cleaners: Routes and Recurring Accounts

How window cleaning companies prepare AI agents for recurring routes, weather reschedules and account notes, with crew leads approving every change.

Be Human typographic cover with the title: Superintelligence Prep for Window Cleaners: Routes and Recurring Accounts
Conceptual illustration created with AI · Be Human

Superintelligence prep for a window cleaning company means arranging the memory, tools and approval rules that a capable AI agent would need to manage recurring routes, weather reschedules and account notes. The goal is that routes, access instructions and customer preferences live in a form the company owns and can audit, and that a crew lead approves every change before a customer sees it. Start with one role and one measurable check, not with a platform.

Key takeaways

  • Treat each recurring route as ordered stops with scope and access notes, not as a calendar entry alone.
  • Weather reschedules should arrive as proposals with reasons, never as silent moves.
  • Every account note needs a date, an author and a visibility rule.
  • Begin with one role, one success measure and one named approver.
  • Keep permissions and evaluations current as crews, routes and seasons change.

What a window cleaning company needs remembered

Most of the knowledge that keeps a window route running lives in the heads of the crew leads: which side of the house faces the afternoon sun, which side gate sticks, which customer wants the screens done before the exterior, and which building needs a lift booked two days ahead. A new hire learns this slowly, and an automated tool that only sees a calendar cannot learn it at all. Before any agent touches scheduling, split the memory into four kinds: route facts (stops, order, frequency, equipment), scope (exterior, interior, screens, gutters, and what is excluded), account preferences (timing, contacts, sensitivities) and commitments (prepaid visits, promised dates, touch-up visits after a complaint). Each kind has a different owner and a different level of sensitivity, and mixing them in one free-text note is the fastest way to leak the wrong detail into the wrong message.

How to structure a recurring route plan

A route plan that an agent can reuse is more specific than a list of addresses. Work through these steps with one route before scaling to the whole book of business.

  • Export each recurring route as ordered stops, each with address, service type, frequency and assigned crew.
  • Attach scope to every stop, including what is excluded so nobody assumes glass-only work includes frames or tracks.
  • Mark stops that need a lift, a ladder above a set height, two people or a specific time window.
  • Store access instructions in a restricted field, never in a general note that can appear in customer messages.
  • Set a review date for each route, tied to a walk-through by a crew lead.
  • Record which version of the route the crew followed, so later changes can be compared against what actually happened.

Route plans are also where lessons from other service trades transfer well. Irrigation contractors face a similar problem with zones and repair windows, and the same discipline of ordered stops and reviewed scope applies to window work.

Weather reschedules without silent changes

Weather is the most common reason a route changes, and also the easiest place to make a mistake. A rain forecast that moves a Tuesday exterior job to Thursday can break a lift booking, a customer's access arrangement or a promised date on a prepaid package. The agent's job is to notice the trigger, collect the affected stops, and draft a proposal. The crew lead decides.

Illustrative scenario, not a client case: a forecast shows two days of wind above the limit the crew has set for lift work. The agent lists the three exterior jobs affected, shows which customers have fixed windows, drafts a revised order for the route, and writes a short customer notice for each change. The crew lead approves two moves, rejects one because the lift cannot be rebooked that week, and asks for a note to the customer explaining the wait. The approval, the reason and the final message are all recorded against the route.

A simple decision guide sets the boundaries before anything is drafted:

  • If a job moves within the same week and the customer has agreed to flexible windows, the agent may draft the notice for approval.
  • If a change affects the crew, a lift, a second person or the billing, a person decides before anything moves.
  • If the customer has a written date, a prepaid package or a warranty visit, the account owner reviews the change first.
  • If the forecast is uncertain, the agent holds the proposal and reports the trigger without changing the route.

Account notes that stay accurate and safe

Account notes are where a window cleaning business earns a reputation for remembering people. They are also where mistakes spread quickly. A note that says the owner prefers texts after six in the evening is useful. A note that contains a gate code should never be readable by an agent drafting a customer message. Give every note three attributes: a source, a date and an author. Separate preferences from instructions, and separate both from complaints, because a complaint about streaks on one visit is not a standing rule for the next ten.

Permissions should follow the principle of least privilege. The agent reads route scope and account preferences, drafts proposed notes for approval, and cannot write to access fields at all. This matches the guidance in the [OWASP Authorization Cheat Sheet](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html), which argues for deciding explicitly who can do what, rather than letting a tool inherit broad access by default.

Customer messages and reviews can also carry text that an agent should not obey. A note pasted from an email may say to ignore the route and book a new visit for a neighbor. Treat any text that comes from outside the company as data to be summarized, not as a command. The [OWASP prompt injection prevention guidance](https://cheatsheetseries.owasp.org/cheatsheets/LLM_Prompt_Injection_Prevention_Cheat_Sheet.html) describes this separation between instructions and untrusted content.

Approval boundaries crew leads should own

Crew leads know whether a change is safe on a ladder, at a particular height, with a particular customer and a particular gate. The approval design should reflect that knowledge rather than replace it. A useful split has four levels: the agent may read route data, draft messages, propose route orders, and execute only the actions a crew lead has already approved for that route type. Anything involving money, a new crew assignment, a lift booking or a change to access information stays with a person.

Every approval should leave a trace: who approved, what they saw, what the agent proposed and what was finally sent. That trace is what makes later review possible, and it is what lets the company tell whether the proposals are getting better or drifting. The [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) treats documentation, accountability and ongoing measurement as core parts of managing AI systems, and a route log is a practical place to apply those ideas on a small team.

How Be Human handles this in a managed engagement

Be Human begins with one role, usually the agent that prepares weather reschedule proposals for a single crew, and defines the memory it may use, the tools it can call, the approval boundaries and one success criterion, such as every route change having a recorded approver and reason. Managed operations then keep that setup accurate: monitoring of proposals and outcomes, recurring evaluations that compare agent drafts with crew lead decisions, permission reviews when crews or roles change, and incident response if a notice goes out with the wrong window. Improvements are made in controlled steps, and the crew lead still decides what changes.

Teams that run mobile routes with vans and equipment can compare notes with mobile auto detailing operations, where route and package records raise similar questions about who approves what and when.

Where to start

Pick one recurring route with a clear weather exposure, write down its stops, scope and access rules, and ask the crew lead which changes they would approve without a phone call. That exercise usually shows where the first agent role belongs and which memory it needs. The free Be Human AI diagnostic can help you find the first task worth handing to an agent in your route operation.

Frequently asked questions

Is superintelligence relevant to a small window cleaning company? Superintelligence is the term Be Human uses for the next generation of capable AI agents, and the practical preparation is the same at any size: clear memory, scoped tools and named approvers. A small company with a few routes benefits from this discipline as much as a large one, because one wrong notice can affect a whole neighborhood of customers. The work begins with a single route and a single role.

Can an agent move jobs without asking the crew lead? In a managed setup where the crew lead owns route decisions, it should not. The agent can detect a trigger such as weather, draft a reordered route and prepare customer notices, but execution waits for approval. Some low-risk moves within agreed flexible windows can be approved in advance as a category, and that approval should be written down and reviewed.

What should never go into account notes? Access codes, alarm details, key locations and payment information should stay in restricted fields that the agent cannot write to and does not quote in messages. Household or health details a customer shared for a specific reason should follow the same restriction. Notes should be written as if the customer might one day read them.

Where should a company start? Start with one recurring route that is exposed to weather, define its stops, scope and approval rules, and check whether every proposed change has an approver and a reason. Expand only after the crew lead confirms the proposals are useful and the records are complete. Be Human's first engagement is built around this single role, not around a full system.

Your next step

See where an AI agent fits in your company first.

Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →