How we work: AI skills ​

Solo founder, plus AI, plus advisers. Some call that "AI founder mode". We call it: humans set the architecture, AI fills in the details, within rails shaped on the result. That way this team delivers at the pace of a much larger team. Hints, not recipes.

In short ​

Coronis is an AI outcomes engine for Belgian businesses. You describe what you need, we estimate the odds and the price, and the result appears on your machine, not on ours. You pay for what worked.

The engine underneath runs on what we call "skills": written rules, limits and examples that tell the AI how each kind of work is done. Today the Coronis stack runs on 19 skills (together about 18,000 words of instructions), on top of a shared library of fixed capabilities. Behind that: one founder with twenty years of IT experience, external advisers, and a decade of production SaaS through Core bv.

A selection of the skills ​

  • Strategic adviser. Dozens of thinking frameworks in one system. The skill picks the three to five most relevant, runs them separately, and weighs the tension between the outcomes. Big choices run through here first before anything is set.
  • Content. The editorial engine behind the guides: fixed article forms, rules for the voice of practice, a source check, and the rule that no piece appears in one language only. Described separately on how we write.
  • Image. From a subject in one line to a few visual options, after which a human picks the best. Fast, with a hand on the choice.
  • Indexing and dashboard. Reads source files (markdown, CSV) locally into a searchable whole. The index is temporary and is rebuilt each time from the source files.

Under each skill sits the real output, not a promise. What you see has gone through these rails and been approved by a human.

How skills grow together ​

Every skill started rough. The first time cost time. After a hundred times, the same skill does in minutes what took half a day at the start.

That is not the AI getting smarter. It is the skill getting sharper: the rules, the limits, the notes on what went wrong. What worked becomes a rule. What broke becomes a limit. What returns three times becomes a new skill. That is why the stack feels faster month after month, without the models underneath changing.

One principle ​

Humans set the architecture. AI fills in the details within that architecture.

That is the whole principle. Architecture here means: the fixed contracts, the editorial rules, the limits we do not tolerate. AI means: proposals and first drafts that have to pass along those rails before they appear anywhere.

Vibes coding versus disciplined AI ​

Much of the software built with AI this year is "vibe coded": prompt the model, ship the output, let the user find the errors. That works for a landing page or a demo video. It does not work for work that a customer or an auditor relies on.

Coronis works the other way round: the rails exist before the AI may write anything. The rails determine what is possible, AI works within them. The outcome of vibes coding is "looks good". The outcome of disciplined AI is "holds up under review".

What AI gave us, and what it did not ​

It did:

  • Content in three languages, instead of Dutch only.
  • A response within the week to a change, instead of per month.
  • A broad system of specialist skills, without a separate subscription for each.

It did not:

  • Remove the need to understand the work. AI does not read the file for you; it asks you to read it.
  • Take away the control. Everything that touches customer data goes through the same gate.
  • Give a free pass for inventions. The check catches it, sometimes the test, sometimes the human.
  • The real customer conversation. That stays human work.

How trust is earned ​

You do not earn trust by saying AI is safe. You earn it through what sits between the AI output and the customer. We do not list the details, but the shape:

  1. Your data stays on your machine. No storage with us, only encrypted transport.
  2. Every substantive claim goes through a source check before publication.
  3. Every output to a customer passes a human approval gate.
  4. What we deliver comes with proof of delivery.

The checks and the gate are the constant. AI is the variable.

Why AI alone does not reach the result ​

AI can prepare, structure and follow up a lot. AI cannot sit with the entrepreneur and decide what really counts for their business. It cannot make the trade-off when a rule clashes with the practice of an SME.

Delivering a result is an AI-supported process with humans in the loop, not an AI-only product. Anyone who sells "AI solves it" without that human step is selling a demo, not a solution. That is not a philosophy. It is how we use Coronis ourselves.

This is not a first time with AI ​

Twenty years in IT and innovation management, including Eurocontrol and Belgian SMEs, shaped my definition of "in production": tested, validated, observable, recoverable. Coronis runs on that same bar. AI speeds up the work. AI does not lower the bar.

And for anyone wondering whether this will last: Core bv has been running production SaaS for more than ten years. Coronis is not an experiment that disappears in twelve months.

Proof ​

(Coronis is in Phase 0: first proven through our own use.)

  • 19 skills in the Coronis stack, plus a shared library of fixed capabilities.
  • About 18,000 words of written instructions that steer the AI.
  • A daily dashboard that brings the work together across the businesses.
  • Guides in the making, each in three languages, with a human at the controls.

P.S. Yes, this piece is also written with AI, with a human in the loop. To the core.