After the November 2025 inflection point, we decided to kick off our brand new AI Track: 7 sessions and growing, a combination of theoretical and hands-on workshops, immediately useful in day-to-day project work. From a Vibe Coding workshop to talks on security and agentic engineering practices, a broad track that moves as fast as the field does — slides, real examples, demos, code-shows, and practical takeaways.
AI Bootcamp
Our typical kickoff is a session to frame the rest of the sessions. Here we took a completely different approach.
The AI landscape moves fast — what was true six months ago is already outdated. Our consultants came in at different points on the frontier: some were already deep into Claude Code 4.5+, others were sharpening workflows they’d built over the past year. To get everyone aligned on the current state of the art, we organized a full-day vibe coding bootcamp so the whole team could see firsthand how much the frontier harnesses have matured in just the last six months.
What we covered
- Claude Code with custom slash commands and skills
- BMAD-Method
- Guardrails
AI & Security
As we’re giving our coding assistants access to our codebases, ticketing systems, email, databases and so on, it is also important to stay aware of the dangers of doing so.
The AI landscape is pretty much the wild west, with high severity CVEs being found all the time and every week another big company falling victim to exfiltration due to AI usage, this is a much needed training.
What we covered
- The Lethal Trifecta: the architectural flaw that makes AI systems fundamentally vulnerable
- Real breaches: zero-click exfiltration, RCE via code comments, SQL injection through a public ticketing system etc
- Indirect Prompt Injection & Jailbreaking
- Defensive strategies
AI Driven Development
A practical tour of how to climb the AI maturity ladder. After our track kickoff with the AI Bootcamp, this session is the follow up: how can you get the most out of your coding agent?
Context Engineering
The central insight is that LLMs are stateless; the harness is where state lives. That context window is your memory budget and context engineering is the art of designing what the model sees.
We talk about memory budgeting, progressive context disclosure, retrieval strategy and eviction policy and finally tool-output shaping.
Compounding Engineering
Each session improves the next one by extending the regular flow of Plan -> Delegate -> Assess with a fourth step “Compound”. You have to avoid having to correct the AI from making the same mistake over and over.
One of the main methods for this is creating and fine-tuning skills, which act as living documentation.
“AI engineering makes you faster today. Compounding makes you faster tomorrow, and each day after.”
— Kieran Klaassen
Harness Engineering
The model can forget instructions, skip steps, or rationalize deviations. Hooks are how we make model deterministic.
We implement backpressure to the agent by applying good engineering principles. The things that we have struggled introducing and enforcing in teams for years are now becoming mandatory to avoid the LLM from freewheeling. TDD, LSP diagnostics, testing, linting, formatting, observability etc they keep the agents in check and make them more productive at the same time.
Upcoming Sessions
MCP, Skills and Extending Your Agent
How we build agents that stay useful past the demo. Model Context Protocol, the Skills Framework, and the design choices that decide whether your agent is a teammate or a chatbot with extra steps.
MCP Servers: Is This How Skynet Started?
A look at the MCP protocol while we build or own MCP Server. What is this RAG thing again? The title is a joke. Hopefully.
Predicting Mental Fatigue Using AI
An applied ML talk, far from the LLM hype cycle: building a model that predicts mental fatigue, what makes the data hard, what the model gets right, and where it should not be trusted. By Pierre.
The Math Behind the AI Curtain
The mathematics underneath the models we use every day — explained the way you’d actually want it explained: enough to know what’s going on, not so much that you wish you’d skipped the room. By Tom.
Coming out of Open Space Day
As we could have expected, AI was a very hot topic at our Open Space Day, with two results for the AI Track:
Our book club’s next read landed on The Alignment Problem: “How do we make machines want what we want?“ A question we’d better have a very clear answer to before (if?) AGI lands.
And three extra sessions, voted up on the day and now being planned:
- Agent Cage Match & Model Bake-Off: Frontier vs Local vs Hybrid. Given the same prompt(s), what is the outcome?
- RAG & Embeddings: Your AI can’t read your docs (yet)
- Text-to-SQL & Semantic Search: Ask your database in plain English
Attending
The AI Track is primarily for itenium consultants. Several sessions are opened up to the wider AI-tenium community — a small, recurring knowledge-sharing group of IT professionals from outside itenium.
AI-tenium is invite-only. If a session looks relevant to you and you’re not already in the loop, reach out.
Why this track works
Interested in our AI Track for your team?
Discover more events
Arrow Tag
We kicked off the season with Arrow Tag in Aalst: quick reflexes, smart tactics, and just the right amount of chaos.
