Articles, guides and field reports on AI agents

EN
Fewer than one in five organisations has a formal governance framework for its autonomous AI agents.
FeaturedGovernance

AI governance: a complete framework for steering your autonomous agents

AI governance for autonomous agents: definition, key stakes, 5 steps to put it in place and the 5 pillars of a framework that holds up in production.

Aurélien Sabaly
October 1, 202621 min→
EN
AI Act fines go up to 7% of global turnover; compliance starts with an inventory of the AI systems a company uses.
Governanceai-actrgpd

AI compliance: AI Act obligations, steps and tools in 2026

Get AI compliance right in 2026: AI Act obligations, risk classification and governance of autonomous agents, with the 2026 and 2027 deadlines covered.

Geoffrey Harrazi
16 min
EN
Header: self-hosting Langfuse v3 in production takes five components to run; the software is free, running it is not.
Guidelangfuseobservabilite

Langfuse: LLM observability and engineering for your AI agents

Langfuse in production: what it traces, what self-hosting v3 costs and what you need to add around it (guardrails, memory, AI Act compliance).

Geoffrey Harrazi
7 min
EN
LangGraph in production: projects rarely fail because of LangGraph itself, but on integration, governance and compliance
Guidelanggraphagents-ia

LangGraph for the enterprise: production-ready AI agents

LangGraph in the enterprise: what it brings compared with CrewAI and AutoGen, its learning curve and what you need to add to reach production.

Geoffrey Harrazi
6 min
EN
Header: out of quota, Antigravity CLI 1.0.3 returns exit 0 and empty output; the Claude Code plugin surfaces the real error.
Newsantigravitygemini-3

Run Gemini 3.5 from inside Claude Code

antigravity-plugin-cc is an open-source Claude Code plugin driving Google's Antigravity CLI (Gemini 3.5): cross-model code review and task delegation.

Geoffrey Harrazi
5 min
EN
Header: fine-tuning costs 15 to 50 times more than a well-architected RAG, with 40 to 60% of its budget spent on data curation.
Comparisonragfine-tuning

RAG vs Fine-tuning: how to choose for your AI agents in 2026

Compare RAG and fine-tuning for your AI agents: real costs, decision criteria and the hybrid RAFT option.

Aurélien Sabaly
11 min
EN
Header: in Idun's test, the small Tiny Aya model translates 2,000 support tickets an hour, versus 137 for a 27B model.
Guidecoheretiny-aya

Building a Multilingual Support Agent with Tiny Aya

We built a multilingual support agent with Cohere Labs' 3.35B Tiny Aya and benchmarked it against models 8-10x its size on technical translation.

Geoffrey Harrazi
14 min
EN
Header: LiteLLM, downloaded over 3 million times a day, was compromised on PyPI on March 24, 2026.
Newssecuritysupply-chain

LiteLLM Supply Chain Attack: What Happened and How to Protect Your Stack

On March 24, 2026, litellm was compromised on PyPI by TeamPCP. What happened, who is affected, and the practices that protect an AI stack.

Geoffrey Harrazi
5 min

Get our guides on the value of AI every month

Field notes on real AI usage in companies: adoption, costs, what it returns and what to do about it. No spam, ever.

1 email per month. No more.

Unsubscribe at any time from the link in every email. How we handle your data: privacy policy