Stacklane is now Vantion LabsLearn more

All work

Quoting and shop-floor software where AI proposes and people decide

6agents, each with one task and a named owner

Software for Dutch sheet-metal and CNC contract suppliers, from a request in the inbox to the invoice: drawings read on intake, costing from what similar parts actually cost, work orders with the drawing at the machine, and six named agents that propose but never act.

A request opened: the drawing on the left, the fields read from its title block in the middle with eight of eleven confirmed, and the conversation about the bend radius on the right

The problem

A sheet-metal supplier gets a request as a mail with a drawing attached, and someone retypes it. They work out a price from memory and from what the last similar part cost, write the quote, chase it, turn it into work orders, follow it round the shop and invoice it. The systems that exist for this are either packaged ERP that assumes you make your own product, or a stack of Excel and a mailbox.

What we built

Datum follows one job the whole way. A mail with a drawing becomes a request under the right customer, and the drawing is read: material, thickness, finish, tolerance and revision come off the title block rather than being retyped. Costing works from what comparable parts actually cost, with quote versions and volume tiers beside it. An accepted quote becomes work orders with a route, and every operation carries the drawing on a tablet at the machine, its own hour booking and its own machine. Outsourced steps are tracked as purchase orders with a return date, so a part at the coater is not a part that has vanished. Quality runs alongside: deviations, an 8D report, and a 3.1 material certificate per batch. The work is talked about where the work is. Every work order and request carries its own conversation with mentions, so a question about a bend radius sits on the part it is about instead of in somebody's inbox, and the hour bookings, the machine states and the late jobs are all on the same record. Six agents sit alongside, one task each: Intake, Opvolging, Orderstatus, Inkoop, Planning and Kwaliteit. Each has a human owner, a monthly credit budget, an autonomy setting and a visible record of what it proposed and how often that was accepted, adjusted or rejected. They read freely and write only proposals. The mailbox and the accounting stay where they are: Microsoft 365 and Google Workspace feed it, invoices and customers go back to Exact Online.

The result

Something you can click through as any of six roles, on a fictional company called Twickelo Plaatwerk, with a shadow run that replays real past email to show what an agent would have proposed before anyone trusts it with a live mailbox. The design, the product specs and the screens are done. The engine is in build, and there are no customers on it yet.

What it produced

agents, each with one task and a named owner

agents, each with one task and a named owner

that can send, price or order on their own

systems it reads from and writes back to

Walk through it

The requests list with its phase per row and an AI column reading drawings, raising questions or reporting an unreadable scan
The AI column says what it did per request. Read, two questions, new mail since reading, or the scan was unreadable. Saying it cannot read something is part of the job.

Want something like this?

Thirty minutes with the founder about one process in your operation. Nothing to prepare.