Meningen die gedekt worden door dingen die ik werkelijk heb opgeleverd.
// De artikelen zijn Engelstalig. Ik vertaal ze niet: een vertaling haalt de scherpte eruit, en het publiek dat ze leest leest Engels.
Making a company AI-ready starts with agreeing where files go. Four boring folder rules, the one mechanism behind them, and the August I found the same dead rule stated as live fact in six of my own files.
"Graph engineering" is three weeks old and already means four different things. The thing underneath it, a code knowledge graph, already has a name and I have been running one across twelve repos for weeks. A January 2026 paper settles the part worth settling: deterministic graphs beat LLM-extracted ones, and mine is honest about which half of it is which.
Months of building an emissions benchmark on public registers, and the model was never the bottleneck. Identity was. Coverage was. Lineage was. What a data foundation has to do before an agent's answers mean anything.
Every project folder gets one file telling any AI tool what the project is, its rules, and where the current truth is written. The filename is AGENTS.md, it is an open standard under the Linux Foundation, and the interesting part is not the benefits. It is the four ways it goes wrong.
Different project, different language, 17 days apart. The mechanism is boring on purpose: a failure written down as a checklist question outlives the fix, and the next reviewer, human or agent, has to answer it.
The bottleneck in working with agents was never the decision. It was being the messenger between two machines. Here is what happened when I removed the relay and kept the decision, including the moment my own approval got thrown out.
One person plus agents is a real operating model in 2026. The design question is not what the agent can do, but which verbs you refuse to delegate: push, merge, declare done.
A portability stance, not a privacy one: keep your project's history, decisions, and state in three plain files you own, and any AI tool can pick up exactly where the last one left off.
A practical follow-up: compacting, context trimming, handover files, and model routing to reduce token overhead in long agentic sessions.
In my agentic AI sessions, 85-98% of token spend was overhead. Here are the changes that actually reduced it.