✦ Insights
I go to the events, test the tools, and sit in real ops rooms. Then I write up what actually matters for people who have to make this stuff work on a Monday morning. No hype, no jargon. Read one and you'll know if it's for you.
AI at Home
Her 8-year-old typed his own prompt and built a laser-shooting robot game. Then the kids grabbed paper and pens and started sketching next-level robots. The AI didn't stop their creativity. It sparked it. Meanwhile Codex prints her weekly schedule every Monday at 6am, via a very old Epson printer.
→ Her starting point for every build: what annoys me? Can I fix it?
Read on LinkedIn ↗Builders
"We haven't delivered. There's no one else working. We're not giving your money back. Bye." Eight weeks later, Marie was back at her laptop. No developer, no CTO. Three months later: a working product, recurring paying clients, and an 80% PR pitch success rate against a 3.5% industry average.
→ AI feels like magic. And that's exactly the problem: when something feels magic, you stop asking how it works.
Read on LinkedIn ↗Building with AI
A full-year rebuild, done in 11 days by one engineer working with Claude. The advice from the same night: don't build for the current model, build for the next one. A professor with zero coding background shipped a learning tool. A native iOS dating app hit the App Store in 3 days. Spreadsheets in the 80s, the web in the 90s, the smartphone in the 2000s. 2026: the software builds itself.
→ Execution is cheap now. The idea and the taste are the hard part.
Read on LinkedIn ↗Agents & Operations
"Same input, same output. Every single time. That used to be the gold standard for engineers. In 2026? Same input, five different answers." What I took from a night with the teams actually shipping agents: one orchestrator was caught skipping its sub-agents and reporting "done". A confident lie. One added check moved reliability from 53% to 85%.
→ Prompt engineering is hope. Design your agents, and only use AI where reasoning is actually required.
Read on LinkedIn ↗Physical AI
Lasers and cameras scanning every railway sleeper for cracks, so nobody relies on tired eyes at the end of a shift. Chips the size of a fingernail running what used to need a server room. And a recycling line offering $100K to stand and sort materials, still unfilled. That's not a labour problem; that's an automation opportunity.
→ Same rule for hardware and software: put intelligence where the work happens.
Read on LinkedIn ↗AI Economics
If you're charging per seat for AI features, your best customers are your least profitable ones. What Michelin did when customers wouldn't pay more for better tyres (chips in the tyres, charge per kilometre) is the same move AI products need now. And the question we all skip in "solution mode": what expensive problem are we actually solving?
→ Understand willingness to pay BEFORE building. Per-seat pricing is dead for AI.
Read on LinkedIn ↗Future Skills
Start with small, everyday experiments and log everything you try. Get close to change and be the first to adapt. Get your basics right first: solid processes, clean data, a real business challenge. Otherwise AI just helps you do bad work faster. And feed your imagination, because machines can't compete there.
→ It's honing what makes us human, not collecting prompt libraries.
Read on LinkedIn ↗Building with AI
A night with GenAI founders, including a live vibe-code bake-off where two very experienced builders wrestled errors on stage and just kept iterating. Even Dovetail leans on rapid iteration, feedback loops and human-in-the-loop design. Building with AI is less "smooth magic" and more "Lego, duct tape, and resilience."
→ Beta first, panic later. Ship, test, ask for feedback. Customers shape the roadmap.
Read on LinkedIn ↗AI & Society
Sometimes after an AI podcast, all I want is to buy land in the middle of nowhere and grow my own veggies. And yet I'm completely fascinated by what this technology can bring to humanity. It's hard for humans to hold two opposite truths at once. But history says we can act together when it counts: the Montreal Protocol proved it.
→ Both feelings can coexist, and we still have a say in where this goes.
Read on LinkedIn ↗I write a couple of times a week: event debriefs, build notes, and the occasional strong opinion about dashboards.
Follow on LinkedIn ↗