Blog

Notes, analyses and data/AI use cases.

Strategy·Mar 24, 2026·20 min

The Hanoutier — How AI Created the Most Formidable Merchant in History

The Hanoutier — the neighborhood store owner who single-handedly manages purchasing, sales, logistics and customer relations — is the best metaphor for the AI solopreneur. Here is why juniors are better positioned than seniors to become the Hanoutiers of the digital era.

hanoutier-ia-solopreneur-revolution

Strategy·Feb 10, 2025·8 min

Data, AI, automation: how to stop confusing your needs

Most projects fail not from lack of technology, but from confusion between what you want to do and what you actually need. A framework for making the right call.

AI·Mar 17, 2026·8 min

AI in Higher Education Summit : what Paris is saying about the future of universities

Sold out, waiting list open. This summit brings together in Paris on March 17 and 18 the decision-makers who will define how 11 million European students learn in five years. What the program reveals, what I expect from each session, and why I will be there with my camera for both days.

AI·Mar 2, 2026·6 min

OpenClaw: the open-source AI agent that finally takes action

Open-source, local-first, and self-hosted, OpenClaw shows the shift from conversational AI to truly executable AI. It does not just answer prompts: it can act, interact with tools, execute real tasks, and raise immediate questions around security, governance, and control.

Strategy·Mar 4, 2026·5 min

From ChatGPT to Claude: beyond the model, a question of architecture and independence

The real issue is not model performance alone, but the architecture, portability, and vendor dependence it creates.

AI·Jan 9, 2025·6 min

RAG without governance is a liability

Access control, audit trails, and evals are mandatory, not nice to have.

Architecture·Jan 3, 2025·5 min

LLMOps essentials: deploy, observe, rollback

What production teams monitor. And what they stop pretending matters.

Architecture·Dec 29, 2024·4 min

MCP and tool connectivity: the new integration layer

Why standardized tool contracts change agent reliability and security.

AI·Dec 21, 2024·5 min

Eval-first AI: test suites before prompts

Golden sets, regressions, and acceptance gates for LLM features.

Strategy·Dec 16, 2024·4 min

AI governance that teams actually follow

Lightweight rules, clear ownership, and auditability without bureaucracy.

Strategy·Dec 10, 2024·5 min

Use-case triage: value × feasibility × adoption risk

A simple scoring model to stop AI theater and ship what sticks.

Strategy·Dec 4, 2024·4 min

From copilots to ops automation: picking the right wedge

Start where friction repeats and trust can be measured.

AI·Mar 10, 2026·7 min

Why 80% of RAG projects fail in production

It's not a model problem. It's an architecture problem, a chunking problem, and a people-confusing-a-POC-with-a-system problem.

Data·Mar 12, 2026·8 min

The real cost of poor data architecture

This isn't abstract technical debt. This is decisions made in 2019 that cost €400k per year in 2025.

AI·Mar 14, 2026·8 min

What CIOs still don't understand about AI agents

Agents are not improved chatbots. They're autonomous processes with memory, tools, and the capacity for action. The CIO who treats them like a standard IT project will be outpaced by their own vendor.

AI·Sep 15, 2025·8 min

An AI agent does what you tell it. The problem is you.

Agent failures in production don't come from the model. They come from ambiguous instructions, undefined scopes, and people who mistake a prompt for a specification.

AI·Oct 20, 2025·9 min

Multi-agent architecture: the real cost nobody calculates

Orchestrating multiple agents multiplies failure points, inference costs, and debugging complexity. Before you build, ask the right question.

Strategy·Nov 5, 2025·7 min

547 billion in AI in 2025: autopsy of an overheated market

The 2025 AI investment numbers are real. What isn't yet real is the business value generated behind them. Anatomy of a gap that will close abruptly.

AI·Aug 12, 2025·8 min

AI agent security: what the CISO doesn't know yet

Prompt injection, privilege escalation, exfiltration via tool use: AI agents open attack vectors your security frameworks haven't integrated yet.

AI·Jul 8, 2025·7 min

An agent without persistent memory: a brain without a nervous system

Agent memory is not an implementation detail. It determines whether your agent can learn, adapt, and be useful beyond the first session.

Strategy·Dec 1, 2025·8 min

Small models vs. big models in enterprise: who wins in 2026

Big models get the attention. Small models make the money. The comparative analysis enterprise teams should do before signing the next license contract.

AI·Jun 15, 2025·6 min

LangSmith and AI observability: from SSL certificate to full tracing

Getting LangSmith to production means solving 40 problems before you get your first useful trace. Here's why AI observability costs more than advertised.

Data·May 20, 2025·7 min

Is your lakehouse ready for AI agents? Probably not.

AI agents need fresh, structured, real-time accessible data. Your lakehouse was designed for batch. That gap is wider than you think.

AI·Apr 10, 2025·8 min

Kill switch, RBAC, scope: three imperatives before putting an agent in production

An agent without an emergency stop, without granular access control, and without a defined action scope is an operational risk. Not a philosophical question.

Strategy·Mar 5, 2025·7 min

EdTech and AI: the announced disappearance of traditional LMS

Moodle, Canvas, Blackboard have dominated EdTech for 20 years. Pedagogical AI agents are making their platform model obsolete. Here's how and why.

AI·Mar 29, 2026·10 min

Cognitive debt: the threat no one sees coming

In 2025 we talked about technical debt. In 2026 the real risk is called cognitive debt — the accumulated cost of poorly managed agent interactions, context loss, and unpredictable behaviors.

AI·Mar 15, 2026·9 min

Agentic AI doesn't scale without clean data

67% of companies have deployed generative AI. Only 20% trust their analytical capabilities. The gap isn't a model problem — it's a data foundation problem.

AI·Mar 8, 2026·11 min

LLM FinOps: what an AI system really costs in production

Between semantic caching, reranking, embedding tokens and LLM calls, the real cost of a production RAG always surprises. A framework for anticipating it.

Strategy·Feb 28, 2026·12 min

AI Act: what it concretely changes for a data architect

The European regulation is live. Risk classification, audit trails, model documentation — here's what changes in architectural practice, not in compliance decks.

Data·Feb 14, 2026·10 min

Lakehouse or Mesh: the real 2026 debate isn't what you think

The lakehouse market grows at 22.9% annually. Data mesh delivered in some contexts and failed in others. A vendor-free analysis to decide based on your actual constraints.

AI·Mar 22, 2026·8 min

MCP: the quiet protocol changing how agents access data

Model Context Protocol flies under the radar but its adoption by Anthropic, OpenAI and enterprise vendors makes it the missing integration layer for production agents.

Data·Mar 1, 2026·7 min

pgvector and PostgreSQL: why the real architectural choice is already made

Snowflake and Databricks have both invested in PostgreSQL. Supabase, Neon and vibe coding have normalized pgvector. This is no longer a debate — it's a convergence to understand.