https://nicolasm.com/en/blog Notes on AI in production and product, by Nicolas Mathieu. en Fri, 13 Nov 2026 08:00:00 +0000 https://nicolasm.com/en/une-regle-ecrite-ne-suffit-pas https://nicolasm.com/en/une-regle-ecrite-ne-suffit-pas Fri, 13 Nov 2026 08:00:00 +0000 My AI follows written rules almost always. For the few that must never be broken, I back them with code: hooks, permissions, no access. How to sort your rules and lock the right ones. https://nicolasm.com/en/faites-de-votre-prompt-une-skill https://nicolasm.com/en/faites-de-votre-prompt-une-skill Fri, 13 Nov 2026 08:00:00 +0000 A reminder of what makes a good prompt, then what to do when you use it daily: why a prompt catalogue gets tedious, and how to turn the prompt into a modular skill. https://nicolasm.com/en/construire-un-processus-agentique-simple https://nicolasm.com/en/construire-un-processus-agentique-simple Fri, 13 Nov 2026 08:00:00 +0000 A lead magnet fills a Google Sheet; every morning a worker emails each new lead three call slots. The rules, the trigger, the message and how I built it with Claude Code. https://nicolasm.com/en/comment-j-utilise-l-ia https://nicolasm.com/en/comment-j-utilise-l-ia Fri, 13 Nov 2026 08:00:00 +0000 With a harness of plain rules, shared memory and checks, I run some 40 projects on my own for about €140 a month. Here's what it runs on and what it costs. https://nicolasm.com/en/shannon-1948-ancetre-des-llm https://nicolasm.com/en/shannon-1948-ancetre-des-llm Sun, 11 Oct 2026 08:00:00 +0000 In 1948, Claude Shannon generated plausible English by opening a book at random. What his experiment says about today’s LLMs, and where to put them to work. https://nicolasm.com/en/markov-1913-ancetre-des-llm https://nicolasm.com/en/markov-1913-ancetre-des-llm Sun, 11 Oct 2026 08:00:00 +0000 In 1913, Andrey Markov hand-counted 20,000 letters of Eugene Onegin. His method is the direct ancestor of next-word prediction in today’s LLMs. https://nicolasm.com/en/turing-1950-test-ancetre-des-llm https://nicolasm.com/en/turing-1950-test-ancetre-des-llm Sun, 11 Oct 2026 08:00:00 +0000 In 1950, Alan Turing proposed the imitation game. In 2025, GPT-4.5 was judged human 73% of the time. Yet seeming human says nothing about being right. https://nicolasm.com/en/lovelace-1843-ancetre-des-llm https://nicolasm.com/en/lovelace-1843-ancetre-des-llm Sun, 11 Oct 2026 08:00:00 +0000 In 1843, Ada Lovelace wrote that the machine originates nothing and does what we order it to do. Turing answered in 1950. What it means for your AI projects. https://nicolasm.com/en/rosenblatt-perceptron-1958-ancetre-des-llm https://nicolasm.com/en/rosenblatt-perceptron-1958-ancetre-des-llm Sun, 11 Oct 2026 08:00:00 +0000 In 1958 The New York Times announced a machine conscious of its own existence. Rosenblatt’s perceptron, its real limits, and what it says about today’s LLMs. https://nicolasm.com/en/eliza-1966-ancetre-des-llm https://nicolasm.com/en/eliza-1966-ancetre-des-llm Sun, 11 Oct 2026 08:00:00 +0000 In 1966, ELIZA imitated a therapist with simple keyword rules, and users confided in it. What the ELIZA effect tells us about AI assistants today. https://nicolasm.com/en/asimov-1991-deux-intelligences https://nicolasm.com/en/asimov-1991-deux-intelligences Sun, 11 Oct 2026 08:00:00 +0000 In his last major interview, Asimov describes AI as an intelligence different from ours. Still the best frame for dividing the work between tool and team. https://nicolasm.com/en/dartmouth-1956-premier-devis-ia-sous-estime https://nicolasm.com/en/dartmouth-1956-premier-devis-ia-sous-estime Sun, 11 Oct 2026 08:00:00 +0000 In 1955, McCarthy, Minsky, Rochester and Shannon asked for $13,500 and one summer to advance AI. What happened, and what it says about AI quotes today. https://nicolasm.com/en/jelinek-ibm-statistique-ancetre-des-llm https://nicolasm.com/en/jelinek-ibm-statistique-ancetre-des-llm Sun, 11 Oct 2026 08:00:00 +0000 In the 1970s, Frederick Jelinek and his IBM team replaced grammar rules with probabilities. The language model at the heart of today’s LLMs starts there. https://nicolasm.com/en/poc-ia-production-pourquoi-ca-echoue https://nicolasm.com/en/poc-ia-production-pourquoi-ca-echoue Fri, 09 Oct 2026 08:00:00 +0000 Why so many AI proofs of concept never reach production, and how to break out: business ownership, measurement, real data, process and compliance. https://nicolasm.com/en/ai-act-classification-annexe-iii-en-pratique https://nicolasm.com/en/ai-act-classification-annexe-iii-en-pratique Fri, 09 Oct 2026 08:00:00 +0000 The AI Act in practice: inventory your AI systems, classify them against Annex III, tell provider from deployer, and plan around the 2025–2028 timeline. https://nicolasm.com/en/mesurer-un-assistant-ia-support-client https://nicolasm.com/en/mesurer-un-assistant-ia-support-client Fri, 09 Oct 2026 08:00:00 +0000 Deflection, intent accuracy, resolution time: how to define, instrument and read the three metrics of an AI support assistant without fooling yourself. https://nicolasm.com/en/comment-doit-etre-un-cv-de-product-manager https://nicolasm.com/en/comment-doit-etre-un-cv-de-product-manager Fri, 01 Mar 2024 08:00:00 +0000 Product manager CV: the formula for a strong line, a before-and-after example, ATS, tailoring to the role, mistakes to avoid, and AI experience in 2026. https://nicolasm.com/en/rasci https://nicolasm.com/en/rasci Tue, 15 Mar 2022 08:00:00 +0000 The RASCI matrix: the five roles, RACI or RASCI, a workshop to build it, a full matrix for an AI system in production and how to keep it up to date. https://nicolasm.com/en/market-fit-definition https://nicolasm.com/en/market-fit-definition Mon, 14 Mar 2022 08:00:00 +0000 Product-market fit: definition, symptoms versus fundamentals, a worked example with numbers, a checklist, mistakes to spot, and what AI changes. https://nicolasm.com/en/concevoir-votre-produit-minimum-viable-mvp https://nicolasm.com/en/concevoir-votre-produit-minimum-viable-mvp Wed, 09 Mar 2022 08:00:00 +0000 MVP: definition, possible forms, a five-step method, a worked example, mistakes to spot, and what AI changes in 2026, revisited in hindsight. https://nicolasm.com/en/backlog-produit-erreurs-a-eviter https://nicolasm.com/en/backlog-produit-erreurs-a-eviter Wed, 09 Mar 2022 08:00:00 +0000 Product backlog: a triage routine for incoming requests, a grooming session step by step, the seven mistakes to avoid, a quick health check and what AI changes. https://nicolasm.com/en/hypotheses-produit-roadmap-et-market-fit https://nicolasm.com/en/hypotheses-produit-roadmap-et-market-fit Wed, 02 Mar 2022 08:00:00 +0000 Product hypotheses: a ready-to-fill hypothesis card, risk-based ordering on a scored example, mistakes to spot and the roadmap that leads to market fit. https://nicolasm.com/en/creer-product-roadmap-agile-asana https://nicolasm.com/en/creer-product-roadmap-agile-asana Sat, 26 Feb 2022 08:00:00 +0000 An agile product roadmap in Asana: time axis, custom fields, a filled-in example item, rituals, mistakes to spot, and the link with Scrum sprints. https://nicolasm.com/en/product-roadmap-comment-faire-pour-la-rediger https://nicolasm.com/en/product-roadmap-comment-faire-pour-la-rediger Fri, 25 Feb 2022 08:00:00 +0000 Product roadmap: who it is for, six steps, prioritisation, an annotated example and how it evolves, mistakes to spot, and where AI projects belong. https://nicolasm.com/en/roadmap-agile-comment-impliquer-equipes https://nicolasm.com/en/roadmap-agile-comment-impliquer-equipes Fri, 25 Feb 2022 08:00:00 +0000 Agile roadmap: eight principles for involving developers and business teams, a worked quarter of planning and the rules that stop it becoming a committee. https://nicolasm.com/en/comment-les-product-manager-etablissent-ils-une-roadmap-agile https://nicolasm.com/en/comment-les-product-manager-etablissent-ils-une-roadmap-agile Tue, 01 Feb 2022 08:00:00 +0000 Agile roadmap: how to build one, a worked quarter-planning example, involving teams without a committee, mistakes to spot early and a 2026 look back.