Writing
Field notes,
not thought leadership.
Everything here comes from operating a real agentic company or building them for clients. Numbers included, failures included. New essay most weeks, and there's RSS if you'd rather not depend on an algorithm.
The useful lesson from Notion's global AI report is not that everyone is behind. It is that most companies are still confusing tool access with operational change.
Singapore will co-fund a remarkable share of a sensible AI project in 2026, if you know which scheme does what, and which windows are closing.
External partners succeed at roughly twice the rate of internal builds, but that statistic, quoted by a partner-side business, deserves your scrutiny. Here's the whole tree.
Most first AI agents fail because of where they were pointed, not how they were built. Here's the scorecard we wish we'd had.
The adoption numbers everyone quotes as a training problem are actually a trust problem, and trust is built in the design, not the rollout comms.
A policy nobody reads protects nobody. Here's the one page that does the job, template included.
Day rates, audit fees, retainers and the grants that halve them, with our own price list included as a data point.
We run our company on agents. Here is the bill, the hours, and the failures, itemised.
The models are fine. The problem is everything your organisation wraps around them, and that's fixable.
When only a hundred-odd vendors out of thousands are the real thing, the buyer's job is mostly fraud detection. Here's the six-question test.
Agents do the legwork of our close across three entities and five currencies, but every journal is human-approved, and nothing leaves the building unsupervised.