Software and AI Agents for Machine Builders
A request for a robot cell comes in. Sales works out which variant fits and which options work together; then the bill of materials, routing and quote are produced. Meanwhile someone looks for the next companies the product could suit and writes to them. In special-purpose machine building, both tasks rest on a few people who also understand the engineering.
Part of this work follows fixed rules; another part looks different in every case. Software for machine builders has to keep the two apart. Variant rules belong in a system that applies them reliably; research, requests and drafts can be taken over by an AI agent when a person checks the result.
What an AI agent can prepare in machine building
- Sales research: read prospective customers' websites for machine park, batch sizes and job ads, explain the fit and draft the outreach.
- Requests: compare an incoming request with the available variants, name missing details and draft the questions.
- Service requests: match a report to the right machine and its history and prepare a reply.
- Reconciliations: compare data between ERP, spreadsheets and project status and report discrepancies.
What this looks like at MAFU-SHERPA
MAFU-SHERPA develops AI-controlled CNC automation for small and medium batch sizes. For its sales team, CERTAINCE built an AI sales assistant that runs in production. It reads manufacturers' websites, scores the fit from 0 to 100 with reasons and drafts an email for the contact's role. A salesperson reviews the draft and sends it from Outlook.
The fixed rules, however, do not sit with the agent. The variant logic lives in the CPQ configurator in Business Central, which produces bills of materials, routings and sales quotes; hundreds of quotes have come out of it. The DATEV export from Business Central is a fixed automation without AI, because the rule is clear. The machine productivity portal shows each customer only the data of its own machines; more than 100 robots send data into it every day. How these projects fit together is told in the digitization of MAFU-SHERPA.
Where the agent stops
The agent does not set a price or commit to a delivery date. It prepares; a person checks and decides. We set how thorough that check is per task, by what a mistake would cost: a research draft needs a quick look, a quote needs an approval.
What the same way of working looks like for requests with drawings is described on AI agents and software for contract manufacturing.
What we build for this is yours: tested, maintainable code. It usually starts with an opportunity assessment that shows which tasks are suitable. If you would rather go through one case directly, let us discuss a concrete workflow.