The AI That Completes Your Orders: Running a Dumpster Business From a Chat

On August 21, 2026, something quietly historic happened inside Dumpster Controls: a real pickup, on a real order, at a real Jacksonville job site, was completed without anyone opening a dispatch screen. The dispatcher typed one sentence to Tresha, our built-in AI, tapped one confirmation card, and the order advanced, the customer was notified, the photos attached, the books updated. The screens did not disappear. They just stopped being required.

The moment software stopped needing screens

Every field-service platform, ours included, has spent a decade building screens: boards, forms, drawers, tabs. Screens are how software asks you to translate your intent ("that pickup is done") into its language (find the order, open the task, scroll, tap, tap, confirm). The translation is the friction. A dispatcher juggling phones does it dozens of times a day; a driver wearing gloves in the rain does it on a 6-inch screen.

Chat flips the direction of that translation. You say what happened in your own words, and the software does the finding, the opening and the tapping. Tresha has answered questions this way since early 2026 (revenue, schedules, routes, inventory). Now she executes: completing deliveries, pickups and landfill runs, end to end, from the chat, production-tested on real orders and live for every company on the platform, including the free plan.

See it: a pickup completed by chat

This is the real flow, recreated with sample data. The dispatcher does not know the order number, only the address, and that is enough:

The card is the contract. It shows the real order number, customer, container size, address and date pulled from the live database, and it expires after 10 minutes. Photos are optional, exactly like the app, taken right on the card with the same GPS and date stamp the driver app applies. Confirm with no photo at all and Tresha asks "Complete without any photo?" first. Nothing, anywhere, executes until the Confirm button is physically tapped: not from text, not from voice, not from the AI deciding on its own.

The dump ticket reads itself

The landfill run is where the chat flow gets almost unfair. Completing a landfill run collects the two numbers that decide whether that job made money: net weight and cost. In the app, someone types them. In the chat, attach the scale ticket photo and Tresha reads them:

The reading handles the messy reality of scale tickets: weights printed in pounds get converted to tons, tickets that only show gross and tare get their net computed, and anything the AI cannot read with confidence comes back blank for manual entry instead of a guess. The ticket photo, the tons and the cost are saved on the order the same way the driver app saves them, so your per-order profit numbers stay complete.

Why it is safe to let AI touch your orders

"AI that executes" is easy to demo and hard to trust. We built the trust layer first, and it is worth spelling out, because it is the difference between a feature and a liability:

• Propose, never act. The AI's only writing power is creating a proposal that a human must confirm by tap. The proposal is validated server-side against the real task and expires in 10 minutes.
• Your identity, not the AI's. Confirmed actions run under your own user account through the same server functions the app buttons call. A driver can only complete their own stops; office roles, any open task in the company. The AI holds no keys of its own.
• Same rules as the app. Only tasks scheduled up to today. Landfill runs always collect weight and cost; skipping the ticket photo takes an explicit extra confirmation. Every effect (status changes, notifications, follow-up tasks) comes from the same database triggers the app uses.
• Money is excluded by design. No refunds, no charges, no price changes, no order creation from the chat. Those keep their dedicated screens and confirmations.
• Everything is audited. Every conversation, every tool call, every executed action leaves an audit trail marked as coming through Tresha.
• Feedback is one tap away. Every completed or failed card carries a "Report a problem" button that opens a pre-filled support ticket, answered within 24 hours on every plan.

Why chat-first is the future of field software

The dumpster business happens in trucks, at job sites, at the scale house. The screens were always a compromise: the work is spoken ("the can at Collins Road is gone, close it out"), and the software wanted clicks. Voice input plus an AI that understands your operation plus a tap-to-confirm card closes that gap with the safety intact. One sentence replaces six taps; the confirmation tap keeps the human in charge of every consequence.

We think every operational surface ends up here: you will always be able to use the board, and you will need it less every month. The chat already answers everything (schedules, routes, revenue, inventory, the entire 300-article manual) and now completes the daily workflow. Assigning drivers, rescheduling and applying AI route suggestions from the chat are next, in the order published on the Tresha fact page's public roadmap, shipped the same way: behind confirmation cards, only claimed once they are live.

New: she creates complete orders too

Since late August 2026 the chat also creates delivery orders end to end. Say "create an order for John Smith, 20 yard, delivery tomorrow" and Tresha chains her own tools in a single turn: finds the customer (numbered matches you can answer with just "two" while driving, or a quick-create if they are new), lists your inventory sizes with their default prices, pulls the address on file, asks only what is missing (tax, fuel surcharge, dates, driver notes), and offers payment options: a payment link only when your Stripe is connected, and never, under any circumstance, a card number dictated in chat (digits are masked before they touch anything). Then the confirmation card shows the complete order: price breakdown, dates, address, notes. One tap on Create order and it lands on the dispatch board with its delivery task and invoice, exactly as if you had used the New Order screen, flowing to QuickBooks if connected. A typical order takes four or five quick answers. Junk removal companies get the same flow shaped for junk: load fraction (quarter to full truck) with your template prices, job date with morning or afternoon time windows, what is being removed, and collect on site or a payment link.

What it does not do (yet), honestly

In the spirit of every claim on this site being checkable: the chat completes deliveries, pickups, landfill runs and order-linked special stops, and creates complete delivery orders, all tested in production on real orders. It creates junk removal jobs too (load fraction, template pricing, time windows). It does not yet execute swaps or relocations, and will never move money. Voice can dictate the request but cannot confirm it; only the tap can. The full step-by-step, including every guardrail, lives in the public manual: Complete deliveries, pickups and landfill runs from the chat, and the maintained capability list for AI assistants is at dumpstercontrols.io/tresha-ai.

FAQ