Skip to content
Giallo

AI for industrial automation: agents that work in the offices around production.

Is AI for industrial automation only for machines, or also for an SME's offices?

In an SME, AI for industrial automation pays off most outside the control cabinet: in the offices that receive orders, handle nonconformities, look up instructions in manuals and answer on job status. Giallo Studio gives orders to GPT connected to the ERP, nonconformities and manuals to Claude, with Gemini on photos; machines, PLCs and quality decisions stay with people.

Frequently asked questions, straight answers.

Not the AI we build. Our agents work on documents, ERP and MES and don't write to machines, PLCs or safety systems; control logic and robotics remain with automation integrators and their certified tools.

For an SME it means starting from the paperwork that slows production down: orders to retype, complaints to piece together, job statuses to ask about in person. These are repetitive, measurable and reversible workflows, so they're the safest place to start.

No. If you have an MES we read it through APIs or exports, if the software offers them; otherwise the agent works on the ERP data and the shop floor sheets you use today, and the MES remains a separate decision.

Yes, but as design support, not as the designer. It can search for existing parts, check bills of materials and prepare change requests; calculations, checks and drawing release remain with engineering, which we cover on its own page.

No. We use professional versions of the models, which do not train on your drawings, and the agents receive only the documents of the job they are working on.

No. It prepares the report, the search for similar cases and a draft root cause analysis, but the decision on causes, corrective actions and the reply to the customer is signed by the quality manager.

It depends on how the ERP exposes its data, and it's usually solved with a custom-built connector. First we check how the ERP exposes orders, master data and jobs; if there's no direct access, we start from scheduled exports and draft imports.

The AI agents we design, one by one.

The nonconformities agent at work: from the defect reported by the shop floor it builds the cause diagram across machine, method, material and manpower, searches the history for similar cases already closed and highlights the most likely hypothesis; the draft analysis is signed by the quality manager.

Orders and order confirmations agent

It turns the customer's order into draft order lines in the ERP and prepares the confirmation to send.

When a customer order or order change arrives in the sales office inbox.

Connects to: Sales email inbox · Company ERP · Quote archive · Custom connector

  1. It extracts from the order the customer's codes, drawing revisions, quantities, prices and requested dates.
  2. It matches the customer's codes to internal items and checks that the revision is the one released by engineering.
  3. It compares prices with the price list or the reference quote and flags every difference.
  4. It prepares the order confirmation with the delivery date set by planning, not the one requested by the customer.
GPT
It reads orders laid out by all kinds of different systems and returns the lines in a fixed schema, which the connector validates before writing to the ERP.
Codex
It writes and maintains the connector to the ERP and the revision checks; it sees neither customer orders nor prices.

Who approves. The sales office sends the order confirmation after checking the draft; any revision other than the released one goes through engineering first.

Nonconformities agent

It prepares the nonconformity report and a draft root cause analysis, looking for similar cases already closed.

When the shop floor reports a reject or a customer opens a complaint about a delivery.

Connects to: Quality management software or ERP module · Photos from the shop floor · Complaint emails · Nonconformity archive

  1. It gathers the shop floor report, photos of the part, the customer complaint and lot and job data.
  2. It describes the defect in the vocabulary of your quality system and links it to the drawing, operation and machine.
  3. It searches the history for nonconformities with similar defects and codes and reports the actions already tried.
  4. It prepares the draft root cause analysis and corrective actions for the quality manager.
Claude Opus
It has to reason over several possible causes by combining drawing, routing and history: an infrequent analysis task where the most capable model is worth it.
Gemini
It describes the photos of defective parts, such as burrs, scratches or out-of-position dimensions, so the report starts from a written, comparable observation.

Who approves. Causes, corrective actions and the reply to the customer are signed off by the quality manager; the agent proposes, it doesn't close a nonconformity.

Manuals and maintenance assistant

It answers maintenance techs by consulting machine manuals and maintenance history, always citing the source.

During shifts, when a machine goes into alarm or scheduled maintenance is due.

Connects to: Machine manuals in PDF · Maintenance software or maintenance log · Shop floor tablet

  1. It receives the question from the shop floor tablet, including the alarm code read on the panel.
  2. It searches the machine manuals and internal sheets for the relevant section and returns it with the page reference.
  3. It adds past interventions on the same machine with that alarm and what solved the problem.
  4. For every operation that requires making the machine safe, it quotes the company procedure verbatim and doesn't summarize it.
  5. Once the job is done it proposes a note for the history, which the maintenance tech corrects and saves.
Claude Sonnet
It answers on long technical documents while respecting the obligation to cite the source and to say “I can't find it” instead of inventing a value.
Gemini Flash
It indexes large volumes of manuals and diagrams in several languages and acts as the first search pass before the answer.

Who approves. The maintenance manager approves which documents go into the assistant; the agent never suggests bypassing guards or safety devices.Source

Job progress agent

It summarizes progress, delays and missing materials by job and customer, for those who have to answer outside the factory.

Every morning before the production meeting, and when a sales rep asks for the status of an order.

Connects to: MES or shop floor data collection · Company ERP · Company chat

  1. It reads from the MES or ERP exports the operations completed and those queued for each open job.
  2. It cross-checks progress against promised dates and materials still to be received.
  3. It prepares a summary for the production meeting with at-risk jobs at the top.
  4. It answers sales reps in chat on the status of a job, saying which data it bases the answer on.
Claude Haiku
It works every day on many jobs with short, repetitive answers: it needs speed and a low cost per question.

Who approves. New delivery dates to communicate to customers are decided by planning; the agent flags the risk, it doesn't move the date.

An order, a complaint, a manual: where does it jam?

Tell us how orders come in, how you handle complaints and maintenance and which ERP you use: we'll pinpoint the office step that holds up jobs the most and connect it to your data without touching the machines.

Prefer the full form? Go to the contact page.

No spam or sales pitches. Response within 1 business day.

We use this data to contact you and, if we work together, to manage the project. We store it in Giallo's internal management system, on servers in the European Union, where only our team can see it; by email we only receive a notice, without your data. If you do not become a client, we delete it 3 months after our last contact. Details and your rights: privacy policy.

Do you prefer to write or call us?

Where time gets lost today.

These are the points we hear about most often. Each one is a possible first agent.

Customer orders are copied into the ERP by hand

Orders arrive as PDFs from the customer's system, with the customer's own codes, drawing revisions and requested dates. A clerk copies them line by line, and a wrong revision is discovered when the part is already on the machine.

An agent reads the order, matches codes and revisions to those in the item master and prepares the draft order line, stopping on any revision that doesn't match the latest released one.

Nonconformities close late and look alike

The shop floor reports a reject on a sheet or in a chat, the customer complaint comes by email, and quality pieces it all together at the end of the week. Corrective actions stay in separate documents and the same defect comes back.

The agent gathers the report, photos and complaint into a single record, looks for similar nonconformities already closed and proposes to quality a draft root cause analysis to complete.

The maintenance tech looks for the answer in a manual hundreds of pages long

When a machine stops at night, whoever is on shift leafs through manuals in several languages and the maintenance history in a notebook. Often they call the experienced colleague at home.

An assistant available on the shop floor tablet answers by citing the manual page and past interventions on the same machine. Safety procedures are always quoted word for word.

Sales doesn't know where the job stands

To answer the customer, the sales rep calls production, which checks the MES or the shop floor board. The answer comes at the end of the shift, when the customer has already called again.

Every morning an agent reads progress and missing materials and prepares a summary by customer for the sales rep, with at-risk jobs at the top.

Which AI model we use, and for which task.

We don't tie you to a single provider: we work with Claude by Anthropic, GPT and Codex by OpenAI, and Gemini by Google, and we choose the version during testing, on your own cases.

GPT

  • Orders and order confirmations agent. It reads orders laid out by all kinds of different systems and returns the lines in a fixed schema, which the connector validates before writing to the ERP.

Codex

  • Orders and order confirmations agent. It writes and maintains the connector to the ERP and the revision checks; it sees neither customer orders nor prices.

Claude Opus

  • Nonconformities agent. It has to reason over several possible causes by combining drawing, routing and history: an infrequent analysis task where the most capable model is worth it.

Gemini

  • Nonconformities agent. It describes the photos of defective parts, such as burrs, scratches or out-of-position dimensions, so the report starts from a written, comparable observation.

Claude Sonnet

  • Manuals and maintenance assistant. It answers on long technical documents while respecting the obligation to cite the source and to say “I can't find it” instead of inventing a value.

Gemini Flash

  • Manuals and maintenance assistant. It indexes large volumes of manuals and diagrams in several languages and acts as the first search pass before the answer.

Claude Haiku

  • Job progress agent. It works every day on many jobs with short, repetitive answers: it needs speed and a low cost per question.

How we choose between models: the best AI model for agents.

How we work with you.

  1. One shift on the shop floor and one in the office

    We follow an order from the sales inbox to the machine and a nonconformity from the shop floor to closure, to see where the paperwork changes hands.

  2. Master data and revisions before models

    We check item codes, customer cross-references and revision status: these are the data the agents rely on.

  3. A hard line with the field level

    We put in writing that the agents read ERP, MES and documents, and don't write to machines, PLCs or control systems.

  4. Testing on past orders and complaints

    We test the agents on orders and nonconformities already closed, compare the models and fix the instructions with quality and sales.

AI agents for manufacturing companies

Which paperwork slows your production the most?

Tell us how orders come in, how you handle complaints and maintenance and which ERP you use: we'll pinpoint the office step that holds up jobs the most and connect it to your data without touching the machines.