This course is currently taught in Dutch only. The page below describes the programme for context. An English cohort may follow later, depending on demand.
AI Agent Sprint
Six sessions to understand, build, and deploy AI agents in your metals operation. Online, in a small cohort, with live demos on your situation.
A structured course for production teams and leadership at metal manufacturers. No demo tricks, no sales pitch, just the real steps to start working with agents on the shop floor.
Course starts
First week of September 2026
Enrolment closes
31 July 2026
Your team is drowning in repetitive work
Your production data finally flows. You have visibility. The MES works.
But the administrative burden didn't disappear, it shifted. Someone still manually checks which orders are late and emails customers. Someone still copies data between systems because they don't talk to each other. Someone still builds the same report every Monday morning. Someone still answers the same supplier questions over and over.
This is work that follows patterns. Predictable. Repetitive. And it consumes hours every week from people who should be solving harder problems.
Planning still takes too long
You have the data, but someone still has to interpret it, check constraints, and build the schedule. Every change means manual rework.
Reports get assembled by hand
Production summaries. Customer updates. KPI dashboards for management. Someone pulls the data, formats it, sends it out. Every time.
Systems don't talk to each other
Your ERP knows about orders. Your MES knows about production. Your email knows about customer requests. Connecting them means someone copying and pasting.
The same questions get answered repeatedly
"When will my order ship?" "What's the status on job 4521?" "Do we have capacity next week?" Someone looks it up and types a response. Again.
You've probably heard about AI automation. Maybe tried ChatGPT for some tasks. But generic AI doesn't know your orders, your machines, your production status. It hallucinates. It guesses. It's not connected to your reality. The missing piece isn't AI, it's AI that understands your factory.
AI that works with your actual data
Imagine an agent that checks your production status every morning, identifies orders at risk of being late, and drafts customer notifications, ready for review before you finish your coffee.
A workflow that receives a customer inquiry, looks up their order in your MES, checks the current operation, and drafts a response with the actual status. Seconds, not minutes.
Planning assistance that understands your capacity, your constraints, your current load, and suggests a schedule instead of waiting for someone to build it manually.
Reports that generate themselves from live production data. No copying. No formatting. Just review and send.
This is what happens when AI connects to structured factory data. Not generic chatbots. Agents that know your operations, your orders, your reality. The administrative burden finally shrinks, not by hiring more people, but by automating the patterns.
This builds on your production data
AI agents are only as good as the data they can access. Without structured, real-time production data, agents guess, and guessing creates more problems than it solves.
The AI Agent Sprint works best when you have:
- MES capabilities in place, work orders, operations, and production status flowing through a system your agents can query.
- A Unified Namespace or similar architecture, structured data that agents can access reliably.
- Completed the MES Accelerator, or have equivalent capabilities already running.
If you don't have production data structured yet, that's where we start. The MES Accelerator builds the foundation. The AI Agent Sprint builds on top of it.
For back-office automation that doesn't require production data (document processing, email workflows, general administrative tasks), we can discuss what's possible with your current systems.
Learn about the MES Accelerator →How the course works
The AI Agent Sprint is a six-session online course that teaches you how agents work, what you need to deploy them in a metal manufacturing operation, and how to build one working agent for your own situation. Each cohort is capped at 10 companies (around 25 participants total as a guideline), so there's room for your questions.
Identify the right tasks
Not everything should be automated. We map your recurring administrative work and identify tasks that are predictable, repetitive, and high-value to automate. The sweet spot: clear patterns, real time savings, low risk if something goes wrong.
Choose the right tools
Microsoft Copilot Studio. N8N workflows. Custom Python scripts. Claude or GPT APIs. The tool depends on your environment, your team's skills, and what you're trying to accomplish. I help you choose what fits, no vendor loyalty.
Connect to your data
This is where most AI projects fail. We design how agents access your production data, your ERP, your documents. Secure. Reliable. No hallucinations because the agent queries real information.
Build working automations
Your team builds with my guidance. Hands-on, not watching from the sidelines. You understand every step so you can maintain and extend it after the program ends.
Test and iterate
First versions rarely work perfectly. We test with real scenarios, identify edge cases, and refine until the automation is reliable enough for daily use.
The rhythm:
- Weekly sessions where we work through design decisions and build together
- Demos showing working automations, not concepts
- Email support between sessions
- Templates and patterns for common manufacturing workflows
Your team does the building. You keep the knowledge. No dependency on me to maintain what we create.
Six sessions, one working agent
Biweekly online sessions via Google Meet. Optional exercises in between. A laptop is handy, no prior knowledge required. By the end your team has a working agent and a clear sense of where to build next.
What are agents, and where is AI really at?
Demystifying agents, LLMs, and what fundamentally changed since 2024. Not the hype, not the doom talk, but the reality for SMB metals. What agents can and can't do, how they differ from chatbots and RPA, and which use cases actually work on the shop floor.
What you actually need for AI in your factory
The foundation: data, processes, people, tools. What needs to be in place before an agent delivers value. Mission on paper, processes mapped, house rules for the agent. The choice between cloud LLMs and self-hosting, and what it really costs per month.
Build your first agent, live
Hands-on: build one working agent together during the session. Connect a tool, write a prompt, see errors and fix them. No demo magic, just the real steps your team will take afterwards.
Agents that know your production data
How to connect an agent to the shop floor. Unified Namespace as eyes and ears, MQTT as backbone, and a memory layer that retains what works. Read, decide, write back to your systems, without the agent hallucinating.
Apps on demand: beyond ERP and MES
Vibe-coding custom shop-floor apps in days instead of months. What remains of your ERP, MES, and QMS when the agent becomes the UI. Selection criteria for agent-friendly software: full APIs, open data, no vendor lock-in.
Your 90-day plan for agent-first
What you do Monday. The order of a first skill, a first agent, a first measurable result. The five biggest pitfalls and how to spot them. How to build internal capability so you don't depend on external consultants.
Get a head start
The agent-first series is live now
The Metaal Connect LIVE series on agent-first is the free warm-up for this sprint. Ten episodes, every two weeks, with live demos. Subscribe to the newsletter so you don't miss an episode.
Watch the livestream seriesAutomations that actually save time
These are examples from metals clients. What you build depends on your specific administrative burden, but these patterns appear frequently.
Late order alerts
Agent checks production status daily, identifies orders at risk of missing delivery dates, and drafts customer notifications for review. Proactive communication without manual checking.
Customer status inquiries
Incoming email asks "where's my order?" Agent looks up the order, checks current operation, and drafts a response with actual status. Review and send in seconds.
Production reporting
Daily or weekly summaries generated automatically from MES data. Output per machine, orders completed, issues flagged. No more Monday morning report assembly.
Planning assistance
Agent reviews incoming orders, current capacity, and constraints. Suggests scheduling options for planner to review and adjust. Decision support, not decision replacement.
Document data extraction
Incoming purchase orders, delivery notes, or invoices get parsed automatically. Key data extracted and matched to your systems. Less manual entry.
Supplier communication
Routine supplier inquiries answered automatically using your order and inventory data. Delivery confirmations, quantity checks, standard questions handled without interruption.
Why this works when other AI projects don't
Most AI automation attempts fail for the same reason: the AI doesn't have access to real data.
Generic chatbots hallucinate because they're guessing. They don't know your orders. They don't know your production status. They don't know your customers.
The AI Agent Sprint connects agents to your actual factory data through your Unified Namespace or MES. When an agent needs to answer "when will order 4521 ship?", it queries your system and gets the real answer.
Grounded in real data
Agents query your production systems, not their training data. Responses based on facts, not guesses.
Context-aware
Agents understand relationships, which parts belong to which order, what operations remain, what's blocking progress. Not just raw data lookups.
Secure access
Agents access data through controlled interfaces. No exposing your entire system to external AI services.
Auditable
You can see what data the agent accessed and why it made specific decisions. No black boxes.
This is the payoff of having structured production data. Without it, AI automation is a party trick. With it, you can automate real work.
Ready to automate the repetitive work?
Let's talk about your administrative burden and whether AI automation makes sense for your situation.
Get in TouchThis is a conversation, not a sales pitch. We'll discuss what you're trying to automate, what data you have available, and whether this program is the right fit.
What's included
- Weekly working sessions (video call)
- Tool selection guidance (Copilot Studio, N8N, custom scripts, AI APIs)
- Data connection architecture
- Templates for common manufacturing workflows
- Hands-on building with your team
- Testing and iteration support
- Email support between sessions
What's not included
- Ongoing maintenance after program ends (your team owns this)
- Custom development beyond the program scope (available separately)
- MES or production data setup (see MES Accelerator if needed)
- On-site visits (available if needed, quoted separately)
What Clients Say
"I highly recommend Luke for his exceptional expertise in digitalization and process optimization. Thanks to his commitment and knowledge, we have made significant strides in improving our efficiency and internal collaboration."
Ing. Laurens Singeling
Singeling Constructie-Lastechniek B.V.
"The professional and goal-oriented approach, combined with extensive technical knowledge, resulted in a stable integration. A valuable consultant."
Yves Hesemans
P&G Metaalwerk
"Our investments in automation projects with Luke lead to shorter lead times and lower cost prices."
Ton Koridon
Koridon Industrieel Plaatbewerking B.V.
"Luke's knowledge of digitization helps us make decisions for our smart factory."
Willem Hoek
SAG Alutech B.V.
Questions you probably have
Do we need technical people on our team?
It depends on what we're building. No-code tools like Copilot Studio or N8N can be managed by non-technical staff with training. More complex integrations benefit from having someone comfortable with APIs or scripting. We assess your team's skills and choose tools accordingly.
What if we don't have MES or production data structured yet?
For production-related automation, structured data is essential. The MES Accelerator builds that foundation. For back-office automation that doesn't touch production (document processing, general email workflows), we can work with your existing systems. Let's discuss what you have and what you're trying to automate.
Will AI agents make mistakes?
Yes, sometimes. That's why we design workflows with human review for anything consequential. Agents draft responses, humans approve them. Agents suggest schedules, planners confirm them. The goal is reducing manual work, not removing human judgment.
What tools will we use?
Depends on your environment and needs. Microsoft Copilot Studio works well for M365 shops. N8N is flexible for complex workflows. Custom Python scripts handle specific integrations. Claude or GPT APIs for language tasks. I help you choose what fits, no vendor loyalty on my end.
What if the automation doesn't work as expected?
That's part of the process. We iterate together, understand what went wrong, and improve it. The goal is knowledge transfer, your team learns how to troubleshoot and extend automations after the program ends.
Can you just build it for us?
Implementation work is available separately. The program itself is advisory, you get the architecture, guidance, and hands-on building experience. Your team keeps the capability.
AI that understands manufacturing
Most AI consultants don't know a press brake from a laser cutter. They'll build you a chatbot that sounds confident and gets everything wrong.
I've spent 15 years on shop floors. I know what data matters, what questions get asked repeatedly, and where administrative burden actually lives in a metals operation.
The AI Agent Sprint isn't a generic AI course. It's AI connected to production reality, because that's the only way it actually works on the shop floor.
I'm independent. No vendor affiliations. If Microsoft Copilot fits your situation, we use that. If N8N is better, we use that. If custom code is the answer, we build that. My recommendation is based on what works for you.
Stop doing work that follows patterns
AI agents that understand your factory. Automation that connects to real data. Knowledge your team keeps.
Enrol nowEach cohort is capped at 10 companies (around 25 participants total as a guideline) so everyone gets room for their own questions and cases. If this round is full, you can register for the next cohort.