Method
An incorrect manual, an inconsistent catalogue, a product value that is difficult to update or an AI agent that responds poorly is often only the visible symptom of a problem that started earlier.
That is why I start from product, information, responsibility, process and context and only move to the final output once their relationships are clear.
The principle
“If information is difficult to find, update, translate, publish or use inside an AI agent, I do not immediately assume that the output is the problem. I look for the point in the chain where that difficulty originates.”
The process
This is not a rigid sequence. In real industrial environments I often go back, validate an assumption and adjust the model. The constant is maintaining coherence between what the product is and what the systems say about it.
The system comes before the document.
I start from the product, the people and the information that already exists. I reconstruct where data originates, who changes it, where it is duplicated and which outputs it must feed.
Make visible what is currently implicit.
I connect content, data, documents and processes to identify redundancies, discontinuities and dependencies that would otherwise remain hidden.
Give knowledge a stable form.
I turn dispersed information into governable models through hierarchies, attributes, taxonomies, editorial rules and clearly defined responsibilities.
Make data, systems and outputs work together.
I try to avoid maintaining the same information in multiple places. ERP, PIM, CMS, documentation and digital channels should share a coherent logic.
AI for orchestration, code for guarantees.
I design deterministic rules and integrated AI agents. AI is used for orchestration and synthesis, while data validation and critical operations remain assigned to verifiable code.
A useful system must outlive the project.
The result should not depend on the person who built it. I document rules and responsibilities so the system can be maintained, extended and improved over time.
Operating principles
If the same information lives in several places, sooner or later those versions will diverge.
A well-designed document cannot compensate for poorly structured information.
Automating an ambiguous process only produces mistakes faster.
Modularity and single-source publishing are not added at the end. They must be built into the content model.
Automation
Python, APIs, JSON, XML and AI agents become useful when the information logic is already clear. At that point, machines can take over checks, transformations, synthesis and repetitive tasks, leaving people responsible for interpretation, exceptions and decisions.
define what should happen
connect rules, data and agents
validate data, quality and exceptions
In short
The final output may be a manual, a catalogue, a PIM, a digital workflow or an AI agent. The method remains the same.
