Writings

2026

• Jul 30, 2026 Agent Memory Is Not a Vector Database
  Enterprise agent memory is not one pile of embeddings. It is layered, governed, permissioned, fresh or stale, auditable, and sometimes deliberately forgotten.
• Jul 23, 2026 When AI Is Asked the Wrong Question
  A retail forecasting problem showed the difference between predicting everything and finding the decision the business actually needed.
• Jul 14, 2026 Private AI Adoption Is Packaging, Not Just Privacy
  Private AI becomes usable when the product tells the truth: what runs locally, what escalates, what hardware can carry, and where the data stays.
• Jul 14, 2026 The Model Can Only Act on Objects the Business Has Defined
  AI readiness starts with business-object semantics: customer, site, device, contract, approval, exception, and the source of truth behind each one.
• Jul 10, 2026 Long Context Still Needs an Owner
  Long-context models may reduce retrieval tricks, but enterprise AI still needs someone accountable for what enters the context, why it was included, and whether it can be trusted.
• Jul 9, 2026 AI Is Usually the Last Step
  The MD Anderson and IBM Watson story is a reminder that AI readiness is often data, integration, workflow, and expectation management before it is a model problem.
• Jul 6, 2026 The Harness Is Where AI Becomes Operational
  A model can reason correctly and still fail the work if the workflow around it cannot enforce completion, handoff, and auditability.
• Jul 6, 2026 Read Is Not Write
  Agent permissions should follow workflow state, not convenience. Read, suggest, execute, and exception handling are different risk classes.
• Jul 2, 2026 Budget for AI Transition Costs
  The hidden cost in AI is often not the model. It is the cost of changing the work around it.
• Jun 29, 2026 When AI Does Not Fit the Work
  The sentence that kills many AI projects is rarely spoken out loud: interesting, but I am not sure where it fits.
• Jun 28, 2026 Model Access Is Operational Control
  The most capable AI models will not always behave like ordinary SaaS. For sensitive workflows, enterprises need to decide where frontier access is enough and where local control is required.
• Jun 25, 2026 The Brutal Test for New Ideas
  A useful idea is not the one the team likes. It is the one whose absence creates a problem for somebody outside the building.
• Jun 19, 2026 AI Adoption Is Change Management
  Giving employees access to AI is not adoption. Adoption starts when ownership, review, risk, and feedback loops change around the work.
• Jun 16, 2026 When AI Becomes a DRP Problem
  The Fable 5 suspension is a reminder that AI continuity is not only a technical problem. For critical workflows, AI now belongs in the DRP conversation.
• Jun 16, 2026 Before the AI Pilot Goes to Production
  An AI pilot is not ready for production just because it works. It is ready when ownership, data access, monitoring, and client exposure are clear before the incident.
• Jun 16, 2026 Agents Need Operational Feedback
  AI agents do not fail only because they lack intelligence. They fail because they see the task, but not the operation around it.
• Jun 9, 2026 Where Frontier Labs Stop
  Frontier labs will keep absorbing generic AI workflows. Durable value sits inside the business, where context, ownership, and accountability cannot be generalized easily.
• Jun 1, 2026 Where AI Work Goes Afterward
  The most important AI record is not the output. It is the interaction that produced it.
• May 24, 2026 The AI Token Is Half a Unit of Measurement
  AI providers can measure what goes in with precision. Most businesses still cannot measure what comes out.
• May 23, 2026 When AI Stops Being an Assistant
  The real bottleneck in enterprise AI is not the model. It is whether the organization and the architecture are ready for systems that can sense, reason, and act.
• May 22, 2026 Loops Find Questions
  When a problem is still taking shape, a good AI loop is often more useful than a polished prompt.
• May 15, 2026 Intelligence Does Not Create Energy. It Unlocks It.
  For centuries, civilization advanced by using intelligence to unlock energy that was already there. AI changes the pattern because it operates on the intelligence side of the loop itself.
• May 13, 2026 The Next Programming Language Will Not Be Written for Us
  As implementation becomes cheap, software engineering shifts away from syntax and toward specification, semantics, and verification.
• May 11, 2026 What Happens When a Company Stops Thinking
  We are starting to automate judgment. Judgment is not a task. It is how we decide what matters.
• May 7, 2026 One Workflow Beats an AI Readiness Deck
  If an SMB wants to know whether it is ready for AI, the fastest way to learn is not a longer assessment. It is one shipped workflow with real owners, real data, and real stakes.
• May 2, 2026 AI Doesn't Run on Tribal Knowledge
  AI didn't create the gaps in your operations. It exposed them.
• Apr 30, 2026 AI Isn't SaaS. The Bill Is Coming.
  GitHub just put a meter on Copilot. Other vendors will follow. The flat AI subscription model and the real economics of inference are starting to collide.
• Apr 29, 2026 Quarante ans. Trois métriques. Zéro qui tient la route.
  Le tokenmaxxing est la dernière version d'un piège très ancien : confondre ce qui est facile à compter avec ce qu'on cherche vraiment à mesurer.
• Apr 29, 2026 Forty Years. Three Metrics. None of Them Work.
  Tokenmaxxing is the latest version of a very old trap: confusing what's easy to count with what you actually want to measure.
• Apr 28, 2026 You Changed Who Worries. That's Not Automation.
  Most AI projects don't replace people. They replace a cheap operator with an expensive supervising engineer.
• Apr 26, 2026 Alice Evolves the Way Code Should
  Most agent systems that claim to improve over time are actually just overwriting themselves. There is a better model -- and it comes from software engineering.
• Apr 20, 2026 Ambient presence beats the better chat box
  ChatGPT's growth is extraordinary, but its flattening may reveal the limit of AI as a destination product rather than a capability embedded inside operational software.
• Apr 11, 2026 Procurement Is Not Transformation
  Confusing deployment with adoption, and adoption with outcome, is the unforgivable sin of enterprise AI.
• Apr 9, 2026 Votre IA a besoin d'un toit. Pas du cloud.
  Le matériel existe. Les modèles existent. Ce qui manquait, c'était un système pensé pour des entreprises qui n'ont pas d'équipe technique.
• Apr 6, 2026 The Roadmap Is Just Paper
  The most interesting AI story I've seen this year didn't come from a lab. It came from a shop floor.
• Apr 2, 2026 Your Private AI Isn't an Assistant. It's a Twin.
  We've been calling it the wrong thing. Private AI isn't a tool that helps employees work faster -- it's a digital twin of the people, processes, and knowledge that make your organization run.
• Mar 31, 2026 The Intelligence Layer
  For 40 years, software waited for you to click. That contract just ended. What comes next isn't a better app -- it's a different relationship with your entire stack.
• Mar 27, 2026 The Job AI Actually Created
  Everyone talks about what AI replaces. Nobody talks about the validation layer it created -- and why you need more senior people, not fewer.
• Mar 26, 2026 Did You Sign an NDA With Your AI?
  You paste client briefs, contracts, and pricing strategies into a cloud AI. The model isn't bound to silence. Your clients don't know you made that decision for them.
• Mar 26, 2026 Cybersecurity Is Not an Annual Event
  Most companies assess their security posture once or twice a year. Private AI changes that equation completely.
• Mar 26, 2026 AI Without Structured Knowledge Is Just Noise at Scale
  The next AI failure won't be a model that can't think. It will be a model that thinks clearly about the wrong things.
• Mar 24, 2026 Where Does Research End and Influence Begin?
  Anthropic just ran the largest qualitative AI study ever. 81,000 people. The findings aren't the story. The method is.
• Mar 23, 2026 The Map Is Gone
  The interface we've used for 40 years is disappearing. And most software is still designed for a world that no longer exists.
• Mar 21, 2026 The Only Thing That Keeps a Company Alive
  Most technology leaders get it backwards. They optimize for risk, for control, for internal excellence. But none of those close deals.
• Mar 19, 2026 AI Is Not a Strategy Problem for Service Companies
  Everyone is telling service firms to become AI companies. That advice will destroy more mid-size firms than AI itself.
• Mar 18, 2026 The Integrator's Dilemma
  Everyone's telling service companies to become AI companies. That's the wrong advice. The winning move is making AI work inside the messy reality you already know.
• Mar 16, 2026 The Prompt Layer Is the New Crown Jewel
  McKinsey's AI platform was reportedly breached in two hours. If they can't secure it, what does that mean for Canadian SMBs using the same tools?
• Mar 13, 2026 Your AI Junior Researcher Works While You Sleep
  Karpathy's AutoResearch runs 100+ ML experiments overnight on a single GPU. The researcher's job just changed.
• Mar 11, 2026 Private AI infrastructure for Canadian SMBs
  Most AI advice to small businesses skips the real question: where does your data live while the AI is doing your work? For Canadian SMBs, the answer shapes trust, compliance, and resilience.
• Mar 8, 2026 The Women Who Stayed
  International Women's Day. Not a celebration post. A recognition of competence so undeniable that the noise becomes irrelevant.
• Mar 8, 2026 Everyone Becomes a Builder
  AI doesn't replace your people. It makes all of them builders. What Anthropic learned internally is what every SMB needs to hear.
• Mar 7, 2026 Why Small Businesses Need Their Own AI
  Cloud AI is convenient until it isn't. For small businesses, owning your AI infrastructure is about control, cost, and common sense.
• Mar 1, 2026 What a Private AI Appliance Actually Looks Like
  A walkthrough of the box that brings AI to businesses that will never hire a developer.
• Mar 1, 2026 The Company Without an IT Team Doesn't Need a Copilot
  They need something that just works.
• Feb 26, 2026 When the Loop Breaks
  What happens when AI removes the consumer from the economic loop?
• Feb 25, 2026 Vision to Trust
  The hardest translation isn't tech to business. It's vision to trust.
• Feb 12, 2026 A place to begin
  Why this site exists, and what it's for.
• Feb 9, 2026 Liberté, Égalité, Fraternité
  Pas comme une devise. Comme un contrat de société.
• Feb 5, 2026 The Gap
  The gap between success and failure isn't talent. It's what you do when no one's watching.
• Jan 13, 2026 The Trust Gap
  Optimism about AI seems less about the technology itself and more about whether people believe the future still has a place for them.
• Jan 2, 2026 Turning 59
  Technology is not the strategy. It is only a lever.

2025

• Nov 14, 2025 The Energy Cost of AI
  Energy is quickly becoming the main constraint in scaling computing resources globally.
• Nov 7, 2025 Bet on the Self-Educators
  You cannot teach people unless they want to learn — and when they want to learn, there is nothing to stop their progress.
• Oct 31, 2025 If There's a Simpler Way, Don't Use AI
  The urge to incorporate AI into enterprise flow is big. I see the pressure and FOMO everywhere.
• Oct 20, 2025 Être critique, c'est bien. Mais à force, on n'avance plus.
  Refuser l'IA, c'est refuser d'écrire le prochain chapitre de notre propre histoire.
• Oct 14, 2025 Le Chat Control ou la tentation de l'Inquisition numérique
  Une société qui veut tout prévenir finit par tout interdire.
• Sep 12, 2025 Garder le cap
  L'aventure ne fait que commencer.
• Aug 12, 2025 Why Bother
  I'd trade a thousand terabytes of knowledge just to feel what it's like to hold a cup of coffee.
• Jul 25, 2025 L'IA ne nous menace pas. Elle nous met au défi.
  Le risque n'est pas l'IA. Le risque est de continuer à piloter nos organisations comme des usines à tâches.
• Jul 15, 2025 Ne lancez pas un projet IA sans préparation
  Un bon cadrage aujourd'hui évite les regrets demain.
• Jul 5, 2025 Le vrai humain ne se trouve pas de l'autre côté de vos écrans
  Si vous cherchez un vrai contact, le regard de vrais humains, vous ne le trouverez plus de l'autre côté de vos écrans.
• Jun 29, 2025 Grossir ou Grandir ?
  Croître n'est pas toujours grandir. La vraie croissance est exigeante. Elle coûte moins mais vaut plus.
• Jun 25, 2025 Décider, c'est s'exposer
  Une stratégie brillante échoue si son exécution est bancale.
• Jun 19, 2025 Attention aux pickpockets numériques
  Ne baissez jamais la garde quand vous êtes en ligne.
• Jun 17, 2025 Train to Be a Plumber
  From the creator of the technology to a troubled alarmist of its consequences — Hinton is following the path of Oppenheimer.
• Jun 13, 2025 What Will Be the Real Achievement in AI Research?
  Those who control these foundation models are the superpowers of tomorrow.
• Jun 9, 2025 AI Cannot Reason — or Can It?
  What really is reasoning? And do we even have a good understanding of reasoning in humans?
• Jun 2, 2025 Quiet Innovation
  Real transformation doesn't always come from the top down — it often starts with people on the ground.
• May 26, 2025 The Never-Depreciating Skills
  What are the long-lasting, never-depreciating skills a knowledge worker needs for a lifetime-resilient career?
• May 14, 2025 Beyond Tokens — the Byte Latent Transformer
  Fixed tokenizers are a source of bias in our models. What if we went back to bytes?
• May 10, 2025 AI, or The Enterprise Engine Upgrade
  We are still at the early times of enterprise AI. Here are the real hurdles companies face.
• May 6, 2025 The AI Divide Is Real — and It's Growing Faster Than We Think
  AI development is concentrated in just a few regions. That concentration risks bigger economic gaps, biased systems, and missed local solutions.
• Apr 14, 2025 Deux visions de l'IA — futur radieux ou gadget temporaire ?
  Les optimistes voient un potentiel énorme. Les sceptiques voient la dépendance. La vérité est entre les deux.
• Apr 9, 2025 Search Is Changing — Optimize for AI
  There is less need to advertise. There is an absolute necessity to appear in organic results to target AI agents.