Chapter 02Agentic AI and ERP

AI agents inside the production flow

An order arrives as a PDF, a delivery route gets planned in a spreadsheet, a quote is priced over the phone. Between those steps, people retype, double-check and chase. An AI agent takes over those handoffs: it reads, queries the systems, prepares the decision and puts it in front of the right person. It also gets maintenance and production working in the same CMMS.

Today

A process built around the ERP

  1. PDFThe order arrives by email.
  2. ERPAn assistant retypes it into the ERP.retyping
  3. SpreadsheetThe schedule is exported to a spreadsheet.copy
  4. EmailThe shop floor gets the schedule by email.handoff
  5. PhoneA slipping lead time gets fixed over the phone.chasing

4 manual handoffs, and nothing flows back: an error corrected today happens again next week.

With an agent

An agentic workflow that learns

  1. Receive

    the order, whatever its format

  2. Read

    customer, part numbers, quantities, requested date

  3. Cross-check

    against the ERP, the stock and the schedule

  4. Propose

    a confirmed lead time, a work order

  5. Approve

    a person accepts or corrects it

  6. Act and log

    the ERP is updated, everything is logged

The feedback loop

Every correction made at approval becomes a rule, with its source and its author. Next time, the agent applies it.

Illustration, using a customer order as the example. The ERP stays in place: what changes is the work around it.

Maintenance and production in the same CMMS

In many plants, maintenance and production each organize their own side: the CMMS (the maintenance management system) on one side, the production schedule on the other, with phone calls and spreadsheets in between. An agent can get them working from the same information.

Workflows at every step of production

A production process is a chain of workflows: at every step, information passes from one person, department or piece of software to the next. Those handoffs are still often done by hand, by email, spreadsheet or phone. That is where an agent has the most to do.

  1. Orders

    receiving, entering, confirming lead times

  2. Planning

    scheduling, assigning workstations and crews

  3. Purchasing

    chasing suppliers, receiving goods, matching invoices

  4. Manufacturing

    work orders, changeovers, instructions at the workstation

  5. Quality

    nonconformities, inspection sheets, customer complaints

  6. Maintenance

    work requests, spare parts, service reports

  7. Shipping

    picking, delivery routes, transport documents

  8. Invoicing

    invoices, reminders, credit notes

Data in every shape

Anatomy of an agentic workflow

The same sequence shows up from one plant to the next. What changes is the data, and who signs off.

  1. Receive

    An order, a message, an alert.

  2. Read

    The agent pulls out what matters, whatever the format.

  3. Cross-check

    It queries the ERP, the stock, the history.

  4. Propose

    It drafts a reply, a schedule, a quote.

  5. Approve

    A person accepts, corrects or rejects it.

  6. Act and log

    The action goes out, and everything is logged.

A harness the company owns

Around the AI model, an agent rests on a harness: its instructions, the business rules, its access to software, the checks and the log of what it does. We build it methodically, documented and versioned, inside your systems, and the company keeps full ownership of it. It can change model or provider without starting over, which makes it resilient. Its teams can also change a workflow as soon as the need shifts, and that is what gives it agility.

Work in progress

At an industrial joinery manufacturer we have worked with since 2025, the open workstreams cover the whole chain:

What we deliver

  • Tools that run on your side

    The goal is agents that work on their own inside your teams, on your data.

  • Partner engineering

    Gilles Babinet leads the strategy; partner engineers build and deploy.

  • Trained teams

    Training days so your people can use the agents, correct them and make them evolve.

  • Support over time

    A year-long engagement rather than a one-off deliverable: workflows change, and the agents keep up.

Let’s talk about your production flows

Tell us where information gets lost between two steps, and we will get back to you to set up a first conversation.