AI agents in the factory: for 8D, OEE and quality work.

I work with manufacturers in Türkiye and with those producing for the EU market. I spent more than twenty years in automotive and textile manufacturing, in quality and operations management, and today I use AI agents in the operations I run. This page answers the questions factories ask most often.

What does an AI agent do in a factory?

An agent is a software assistant that works by set rules. It reads records, prepares drafts and sends reminders. In a factory it saves the most time on:

• 8D drafts for D2 (problem description), D3 (containment) and the timeline
• A daily OEE summary and the biggest losses, from downtime, scrap and shift records
• Reminders for supplier information requests, analysis results and due dates
• Finding similar past complaints and FMEA lines and putting them in front of the engineer
• Shift handover notes and the weekly quality meeting summary

Does the agent write the 8D report by itself?

No. The agent prepares a draft; verifying the root cause, deciding on actions and talking to the customer stay with the quality team. Root causes are found on the shop floor, not at a desk. The agent does not replace that work; it frees up the team's time for it. I explain why in Why should 8D decisions stay with people?

What does the agent do for OEE tracking?

It reads the records you already have: machine data, MES, Excel or paper downtime forms. It calculates availability, performance and quality and summarises the three biggest losses and changes every morning. No new hardware is needed to start. First we make your OEE definitions clear; automating a badly defined OEE only makes the mistake faster.

Does it work with our existing systems? What if we have no MES or ERP?

Yes. The setup adapts to the systems you have: ERPs such as SAP, Logo or Canias, MES, SCADA and machine data where available. We respect the layers between production and business systems (the ISA-95 approach) and connect to the shop-floor network with read-only access, following OT security principles (IEC 62443). I am not tied to any software or hardware brand; I recommend what fits the job.

No MES or ERP is fine too. Most first pilots start with Excel sheets, shared folders and email. If a system change is needed, we discuss it after seeing results.

Where does our data stay?

The agent only accesses the data that job needs. Before setup we write down which data goes where, where it is stored and how access is closed. If your customer contracts or Turkish data protection law (KVKK) do not allow it, that data is not given to the agent. Everything the agent does is logged.

Why lean first, then AI?

Because automating waste makes the waste bigger. Before handing an unnecessary approval step or a repeated report to an agent, we check whether that step is needed at all. It is a habit from my years with the Toyota Production System: see the shop floor first, improve second, automate last.

How do we work in a factory?

We follow four stages, and you decide at the end of each one. The goal is not to deliver software but a measurable improvement:

• 1. On-site review (1–2 days): I look at the work on the line and in the office (gemba). I map losses, repetitive tasks and where data lives (ERP, MES, SCADA, Excel). You get a short written report with the top three opportunities.
• 2. Written target and scope: We pick one line, product family or process. We record the baseline of the indicators we will measure and write down the target and timeline up front.
• 3. Pilot and on-site follow-up (4–8 weeks): The agent first produces drafts only. We measure every week and adjust until we approach the target. At the end of the pilot I give management a summary of effort spent and time gained.
• 4. Roll-out and standard: What works moves to the next line or plant. With standard work, short training and clear ownership, your team runs it on its own.

Which indicators measure success?

No pilot starts without a baseline. Depending on the task we choose a few of these:

• OEE: overall equipment effectiveness from availability, performance and quality; downtime by cause.
• Scrap rate and first-time-right (FTR) rate.
• Quality processes: 8D closure time, repeat complaints, customer PPM.
• Delivery: on-time in-full (OTIF), supplier delays.
• Effort: engineer hours spent on reports, follow-up and correspondence.
• Energy: energy per unit produced (in line with ISO 50001).

Which use cases do you cover?

• OEE and downtime analysis, shift and management reports.
• Quality: 8D, FMEA and control plan drafts; early warnings from reject and complaint data.
• Maintenance: recurring patterns in breakdown and maintenance records; a first step toward predictive maintenance where machine data exists.
• Planning: from order to material needs, alerts for shortages and risky dates.
• Sourcing: supplier follow-up, certificate and test report checks.
• Energy: consumption tracking and deviation alerts.
• For vision projects that need cameras and dedicated hardware, I help you choose and evaluate the right solution, independent of any vendor.

How long does it take and what does it cost?

A pilot usually takes four to six weeks. There are no fixed prices on the site; we agree the scope in the first meeting and I send a written proposal. I work as a boutique practice: you deal with me directly and the work scales to your budget.

Who do you work with?

Manufacturers whose quality and supply chain workload is growing, especially automotive suppliers (IATF 16949), defence and heavy industry. I work with plants in Türkiye and the European Union remotely, and on site when needed.

What experience is this based on?

Hyundai and Ford programmes and process improvement at Assan Hanil, the Toyota Production System and operational excellence at TB Sewtech (Toyota Boshoku), and setting up a factory from scratch in Bangladesh. Today I run operations at INVAMAR and use AI agents myself in finance, HR, operations and business development. Details are in the case studies and on the About page.