AI for junk removal companies: photo quotes, routes and disposal records.

A junk removal job has an odd shape. It is sold from pictures, priced again when someone stands in front of the pile, and finished somewhere the customer never sees: a transfer station, a scrap yard, a charity's loading door. Lose track of any of those three places and the company loses money on the quote, an argument at the curb, or the paper trail that shows where a customer's belongings went.

Supervised AI assistants fit the desk work around those places: drafting a volume range and an item list from the photos the owner passes them, flagging items that need a decision, proposing the order of the day's stops, and filing every receipt and statement against its job. Pricing, the hazard call, the donation choice and any signature need someone standing in front of the pile, which is why each one is the owner's.

Take a company that prices by how much of the truck a load will fill, an eighth, a quarter, half or a full load, with extra charges for items that cost more to get rid of. Its estimate is a judgment about volume, and the customer's photos are the only evidence it has before the visit. Photos flatten depth, hide whatever sits behind the first row of boxes and never show the stairs. An honest estimate made from them is a range with its assumptions written beside it, not a number.

That is the line this article holds throughout. An assistant can turn photos into a draft range and a list of what it could and could not see. The figure the customer agrees to is set by the owner, or by whoever the owner trusts to price on site, after a walk-through and before anything is loaded. The draft exists to make that conversation shorter and to leave a record of what the customer was told in advance.

Customers send pictures however they like, so the photos reach the assistant through the owner: attached to the owner's conversation with the Bot, which saves them into that inquiry's folder on its computer. In Aevrion's pattern this is an Estimate Bot, one the owner creates in Grok Bot with a single job. It reads the photos for one inquiry, lists the visible items, places the load in a truck-fraction range using the owner's written pricing table, and adds a line for every assumption: boxes may continue behind the shelving; access looks like one flight of stairs, not confirmed. It never quotes a figure the table does not support, and it never sends anything to the customer.

Grok Bot's documentation describes a Bot as a durable teammate with a name, a job and a conversation of its own, and it advises a separate Bot only where the work has its own goal, tools, approval boundary or schedule. The desktop composer accepts up to six attachments at a time, with images up to 25 MB each, so a longer set of inquiry photos goes over more than one message. If another Bot needs to look at the same pictures, the documentation is specific: handoffs inside a group chat are currently text-only, so the Estimate Bot sends an image directly to the Bot that has to inspect it. The pricing table lives in the business's own files, because the documentation is clear that a Bot's memory does not replace an authoritative source.

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 walk-through is where a photo estimate is confirmed or corrected. In this illustration, a two-car garage clear-out arrives as four photos: a treadmill, a wall of boxes, an old chest of drawers and, in one corner, a dehumidifier. The Estimate Bot's draft puts the load in a half-truck range, lists the dehumidifier as a refrigerant item, and notes that the far wall is out of shot. On site the owner finds the boxes run three deep, with a shelf of old paint tins behind them. The table shows how each line of the draft is used.

Two things never happen in that exchange. The draft is never shown to the customer as a price, and the Bot never revises the agreed figure afterwards. If the kerbside price moved a long way from the range, the owner can tell the Estimate Bot why, and later drafts carry that reason as a new assumption; the decision itself stays with the person who stood in the garage.

Every junk removal company needs a written answer to what it will not take, and that answer belongs to the owner. The assistant's part is narrow. When a photo, or a crew note the owner passes on, shows paint, solvents, pesticides, pool chemicals, fuel cans, batteries or old building material that might contain asbestos, the item gets a flag and a question for the owner. The Bot does not say whether something is hazardous, how to handle it or how to pack it; those are judgments for the owner and, where needed, the local authority or a specialist the owner chooses.

The flag points outwards for a reason. EPA's household hazardous waste page says products such as paints, cleaners, oils, batteries and pesticides can contain hazardous ingredients and need special care at disposal, that household hazardous waste is regulated at state and local level, and that residents should ask their local environmental, health or solid waste agency about collection (as checked on 16 October 2026). A message the Bot drafts for the owner can therefore point a household towards its local collection program instead of offering handling advice the company is not placed to give.

Routing in this trade is not just customer to customer. Each load has to be emptied before the truck can take the next, and where it goes depends on what is in it: mixed debris to a transfer station, furniture in good condition to a charity that has agreed to take it, metal and appliances to a yard that accepts them. In Aevrion's pattern a Dispatch Bot drafts the next day's order of stops from the booked jobs, each estimate's size and a facility sheet the owner keeps, listing every site's hours, accepted items and rules. The owner moves stops around, confirms the order and drives.

Write the facility sheet down before building the Dispatch Bot, because a Bot can plan only around rules it can read: a transfer station that closes early on Saturdays, a charity dock that takes furniture but not mattresses, a yard that wants appliances kept apart from scrap. Those examples are illustrations; each site's own published rules are the record, and the owner checks them. Reading bookings from the company's software is a custom integration or supervised browser use where the vendor permits it, subject to testing. The guide to a one-truck clean-out day follows a single day through those legs, estate house included.

The last stop produces the record that matters if anyone later asks where a customer's things ended up: a weigh ticket or receipt from the transfer station, a slip from the charity, a ticket from the recycler or yard. A Records Bot, the third in Aevrion's pattern, files each one against its job and lists any job with a load that has no paper. It does not create a receipt that was never issued, and it never tells a customer that a donation is deductible; that question belongs to the customer and their own adviser.

Two rules from outside the company bear on some of that paper, and one guide in this series covers both. In the United States, EPA's safe disposal requirements under Section 608 make whoever takes the last step in disposing of a household refrigerator, freezer or window air conditioner, a scrap recycler or landfill for instance, responsible for refrigerant recovery, or for holding a signed statement or contract when the refrigerant was removed earlier (as checked on 16 October 2026). In England, gov.uk says a business that transports waste must register as a waste carrier, and that failing to register can bring an unlimited fine (as checked on 16 October 2026). The guide to disposal records for fridges, e-waste and transfer notes sets out what each rule asks of a hauler.

Aevrion's starting roster for a junk removal company is three Bots: Estimate, Dispatch and Records. Each is a Bot the owner creates and describes in operational terms, with its standing limits given as lasting rules to remember: never send, never price, never sign. That follows the documentation's advice to start with the smallest useful roster and with read-and-prepare work, adding approved actions or routines only once the results hold up. Customer messages the Bots prepare, such as a booking confirmation or a note that paint tins will not be collected, appear in the owner's conversation as editable email drafts that the owner sends or discards.

The three share one persistent cloud computer, whose files, browser sessions and saved logins are open to every Bot on the account, and separate Bots are not a security boundary. A login to booking software or to a facility's account portal is therefore available to all three, and the owner decides which logins belong there at all; work that needs its own credentials belongs under a separate user account. Bots can use many browser-based tools, but a site may refuse automation, and a person signs in, enters codes and makes payments. No Bot in this design takes calls or texts; the phone stays with the owner.

Take a month of finished jobs and put each photo estimate next to the figure agreed at the walk-through. The jobs where the two differ most show what the photos keep missing, which is what an Estimate Bot's assumption lines should be asking about. Then check the same jobs for a receipt against every load; each gap is a job the company could not account for if asked.

Those mismatched jobs are worth keeping in view while rating photo pricing in Aevrion's workflow automation readiness assessment, which is free and needs no sign-up. It rates a single workflow against twelve fixed dimensions, from repeatability and rule clarity to data availability and security and compliance readiness, and straight away returns a score with one of four results, the blockers it found and a next step. A pricing table that exists only in the owner's head shows up as low rule clarity, and a low score there caps the result. When the photo, route and records work is ready to design, a scoping conversation through the contact page is where to plan the three Bots and test any connection to the company's booking software against what its vendor allows.

1. A junk removal job is sold from photos, priced at the walk-through and closed by a disposal receipt, and each step leaves paper.
2. An Estimate Bot drafts a truck-fraction range with its assumptions listed; the owner sets the price on site before loading.
3. Paint, chemicals and possible asbestos are flagged for the owner to refuse or refer; no Bot judges whether an item is hazardous.
4. A Dispatch Bot drafts each day's stops from a facility sheet the owner keeps for transfer stations, charity docks and yards.
5. A Records Bot files every receipt, slip and statement against its job and lists loads with no paper; it never invents a record.
6. Refrigerant appliances fall under EPA's final-disposer rule in the US, and in England a business that moves waste must register as a carrier.