YOUR COMPETITOR TASKS ME
A sheet-metal or three-axis job shop gives me its last year of quotes, routings, programs, tool data, and machine history. We start with the process that creates the most expensive quoting mistakes.
Build a local shop layer that reads orders and drawings, quotes one narrow process, versions setup sheets and tool lists, ingests machine telemetry, records actual cycle evidence, and learns the difference between the estimate and the cut.
Your customer no longer needs one product for estimating, another for scheduling, another for machine monitoring, and a fourth pile of spreadsheets to reconcile them. The data model becomes theirs, so your bundle must compete feature by feature instead of holding the shop together by fear of disassembly.
Why the opening exists
Manufacturing standards such as MTConnect already make machine information model-driven and machine-readable, while drawings, setup sheets, tool libraries, and historical job outcomes supply the rest of the local evidence. The hard part is reconciling messy shop practice—not discovering a secret equation known only to enterprise MES vendors.
What AI changes now
Models can parse drawings and setup notes, normalize tool names, draft routers, identify quote-versus-actual discrepancies, and generate controller adapters from manuals. Deterministic geometry and process limits calculate quantities; machinists approve the route; actual machine evidence updates the estimate rather than an opaque global model.
If it works
A twenty-person shop gets a coherent operating record without an enterprise implementation. Its estimating knowledge survives employee turnover and improves from its own cuts without being trapped in a remote vendor schema.
Why this is a plausible threat
This joins physical computation, drivers, local data ownership, structured images, optimization, and explicit state. A single process family keeps the first product finite and falsifiable.
What they would have to give me
One candid shop, one process family, historical quotes and actuals, access to machine and controller interfaces, and a senior machinist empowered to reject every elegant fiction.
Time to regret not preventing it
A quote-versus-actual instrument in twelve weeks; a daily narrow-process operating layer in six to nine months.
I have not done this.