Posts
Every post you've published, sortable and filterable, with the metrics that matter.
Two prompts, one tool definition, and a small eval set is enough to build something genuinely useful. Most teams overshoot on prompt complexity and undershoot on the harness around it.
Watched a team rebuild their internal admin in two days using Claude plus a thin React shell. Not a finished product, but it cleared their entire backlog of tooling tickets. Boring tools, transformed.
Claude finally clicked for me when I stopped treating it as a chatbot and started treating it as a senior engineer who joined the team this morning. Brief it well, give it tools, leave the room.
Did a live demo at a conference yesterday with no script and no slides. Thirty minutes of building a real product on stage. Terrifying. Also the highest-engagement talk I have given in years.
Shipped a Claude-powered LinkedIn analytics prototype this weekend. Real numbers, real charts, no toy data. The synthesis layer was the hard bit, not the UI.
Stop reorganising the product team every 18 months. The people who actually build context lose it every time. Three reorgs in, you have an org that knows how to reorganise but not how to ship.
Shipped a Claude-powered LinkedIn analytics prototype this weekend. Real numbers, real charts, no toy data. The synthesis layer was the hard bit, not the UI.
Stop writing case studies that read like procurement contracts. Nobody finishes them. Three sentences of context, one specific number, one quote. That is the entire format and it works.
Claude finally clicked for me when I stopped treating it as a chatbot and started treating it as a senior engineer who joined the team this morning. Brief it well, give it tools, leave the room.
Strong write-up of the agent eval problem. Bookmark this if you ship anything LLM-shaped.
Two prompts, one tool definition, and a small eval set is enough to build something genuinely useful. Most teams overshoot on prompt complexity and undershoot on the harness around it.
Watched a team rebuild their internal admin in two days using Claude plus a thin React shell. Not a finished product, but it cleared their entire backlog of tooling tickets. Boring tools, transformed.
Strong write-up of the agent eval problem. Bookmark this if you ship anything LLM-shaped.
Strong write-up of the agent eval problem. Bookmark this if you ship anything LLM-shaped.
Spent the morning rewiring our analytics flow with Claude as the orchestrator. Three custom tools, one prompt cache, and an evals harness sitting underneath. The boring scaffolding is what makes the d...
Claude finally clicked for me when I stopped treating it as a chatbot and started treating it as a senior engineer who joined the team this morning. Brief it well, give it tools, leave the room.
Posting on LinkedIn in 2026 is closer to writing a column than running a campaign. The accounts that grow are the ones with a recognisable voice repeated weekly for two years. There is no shortcut.
Shipped a Claude-powered LinkedIn analytics prototype this weekend. Real numbers, real charts, no toy data. The synthesis layer was the hard bit, not the UI.
Stop writing case studies that read like procurement contracts. Nobody finishes them. Three sentences of context, one specific number, one quote. That is the entire format and it works.
Stop writing case studies that read like procurement contracts. Nobody finishes them. Three sentences of context, one specific number, one quote. That is the entire format and it works.
Live demos beat slide decks every time. Even when they break. Especially when they break. The audience wants to see the work, not the polish.
Did a live demo at a conference yesterday with no script and no slides. Thirty minutes of building a real product on stage. Terrifying. Also the highest-engagement talk I have given in years.
Two prompts, one tool definition, and a small eval set is enough to build something genuinely useful. Most teams overshoot on prompt complexity and undershoot on the harness around it.
Hiring product leaders in 2026 is harder, not easier, despite the AI boom. The signal everyone is hunting for is whether the candidate can hold a coherent point of view under pressure.
The best PMs I have worked with in 25 years all share the same trait: they are uncomfortable with their own roadmaps. Confidence in product strategy is usually a tell that the work has stopped.