Insights

What I'm learning, shared.

No content for content's sake. Honest notes drawn from real AI systems in production: RAG, agents, costs, on-prem. I write up what's been useful to me, to save you time.

FeaturedAI & automationJuly 2026

MCP, the standard that plugs AI into your tools

An acronym you will run into everywhere in 2026. MCP is the standard socket between AI and your software. What it changes for your budget, your freedom of choice and your security, without the jargon.

Read the article

7 min read

All articles

Quick reads, ideas that hold up.

RAG, agents, LLM costs, production reliability: what I actually see in the field, no fluff.

CostsJuly 2026

How much does an AI agent cost?

There is no catalog price for a custom agent: here is what really drives the budget, from build to run, and the five questions to ask before you sign.

AgentsJuly 2026

What is an AI agent? (and how it differs from a chatbot)

A chatbot answers, an agent acts: the difference in plain words, so you can tell whether your need truly calls for an agent.

RAGJuly 2026

RAG or fine-tuning: which one adapts AI to your data?

RAG or fine-tuning is not a tool choice but a problem choice: memory (facts that move) or method (the form and style).

ComplianceJuly 2026

The EU AI Act for businesses: what actually applies to you

A non-legal, anti-panic decoder of the EU AI Act: who is concerned, the risk-tier logic, the real timeline, and the first useful moves you can make tomorrow.

AI & automationJuly 2026

7 AI automations that pay for themselves fast

Not all automations are equal. Here are the seven that almost always pay off, in the order I put them in place, and two I advise you against.

StrategyJuly 2026

How to evaluate an AI provider (before you sign)

In 2026 everyone 'does AI': here is the checklist I would use, as a buyer, to sort a provider before signing.

AI & automationJuly 2026

n8n, Make or Zapier: which one for your company

All three do the same job on paper. The real gap is the cost when volume rises and where your data ends up. Here is how I decide.

SecurityJuly 2026

Prompt injection explained: the flaw in AI wired to your tools

A hidden instruction inside a document or web page can hijack an AI that has hands on your tools: here is how prompt injection works, and what actually contains it.

AI & automationJuly 2026

Extracting your invoices and contracts: why vision models replaced OCR

For years, reading a document automatically meant stacking character recognition and brittle rules. That is no longer the case.

ReliabilityJuly 2026

Why AI Makes Things Up (and How to Reduce Errors)

An AI produces plausible text, not necessarily true text: here is why it makes up answers, and the three concrete levers that bring errors down.

AI strategy & costsJuly 2026

AI usage policy: framing shadow AI in 2 pages

Your teams already use AI, often on unapproved tools. Here is the short policy I set to protect your data without blocking anyone.

StrategyJune 2026

Build, buy, or bring someone in: deciding on AI

Buy SaaS, hire a team, or bring in an external specialist: three routes to an AI capability, and a simple grid for picking the right one.

StrategyJune 2026

GPT, Claude or open-source: how to choose your model in the enterprise

The best model is rarely the real question: here is the framework I use on client work to choose between GPT, Claude and open-source, and to keep that choice reversible.

AI & automationJune 2026

Is your data ready for AI?

Data preparation is the real first step of an AI project, and the leading cause of silent failure. Here is how to tell whether yours holds up.

SovereigntyJune 2026

Sovereign AI: what it actually means, and when it matters

Behind the sales pitch, three concrete questions (where your data lives, under which law, who can access it) that tell you when sovereignty is a real constraint and when it is just comfort.

StrategyJune 2026

From POC to production: getting an AI project out of purgatory

The model answers well, everyone claps, and six months later the project sits idle: here is why so many AI POCs never ship, and how to scope yours to be production-ready from the start.

ArchitectureJune 2026

MCP vs API: what's actually different, and why it matters

You already have an API, so why another protocol? Because an endpoint that exists and a model that knows when to call it are not the same thing.

AI strategy & costsJune 2026

Cut your LLM bill by 30 to 70%

Caching, model routing, shorter prompts, batching: the levers I actually pull to slash the cost of an AI product without compromising answer quality.

AI & automationJune 2026

Enterprise RAG: where to start

Before plugging an LLM into your documents, you have to clean, chunk and index. That is where the real work is, not in the model.

AI & automationMay 2026

Automate without breaking everything

An automated workflow that fails silently costs more than a manual task. How to put guardrails in place from day one.

AI strategy & costsApril 2026

Fine-tune or prompt

Fine-tuning is tempting, but solid prompting is enough nine times out of ten. The decision framework I run through before spending a dime.

AI & automationMarch 2026

Measuring an AI agent in production

Without continuous evaluation, an agent drifts without warning. The metrics that truly matter, and the ones that give you false comfort.

AI strategy & costsFebruary 2026

Reading an API bill without panicking

Tokens, context, models: decoding an LLM bill line by line to see where the money actually goes.

AI & automationFebruary 2026

Connecting AI to your existing tools

No need to replace anything. How to layer useful AI onto the CRM, support desk or back office you already have.

Stay in the loop

Got something on your mind?

I write what I wish I had read before diving in. If a question has you stuck, ask it on a call: 15 free minutes, the answer will often feed the next article, and at the very least you'll walk away with a straight opinion.

What's next

Enough reading: let's get to your project.

RAG, agents, automation, or simply scoping your AI costs: the first call is free, no strings attached.

Contact

Ready to go from demo to production?

Reply within 24 hours · first conversation free, no strings attached.