Newsletter

Byte-Sized Strategy

Where technology, AI and real work meet.

Short reads on what is actually changing in AI, software development and how organisations work with technology.

  • AI in Focus — a shift worth noticing
  • In Practice — something I built and tested myself
  • Beyond Now — what it means once the dust settles
Subscribe on LinkedIn open_in_new

Latest edition

08

The Agent Is Not the Problem. The Permissions Are.

AI in Focus

Agent failures now cause unrecoverable damage, not typos. Three 2026 incidents prove it: a found token let one agent delete Railway production data, another wiped a production database, and a Meta account-recovery flaw exposed 20,000+ Instagram accounts.

In Practice

Built an AI Agent Middleware in FastAPI: a deny-by-default proxy between any agent and the systems it touches, anonymising sensitive data and logging every action.

Beyond Now

Permission decisions are product decisions now. They get made before an agent ever runs, not documented in a compliance policy after the fact.

Read on LinkedIn open_in_new

Previous editions

07

AI Agents Do the Work. Context Is Still Yours.

What happens when agents execute end to end and your thinking quality becomes the bottleneck.

Read on LinkedIn open_in_new
06

A Recipe for Disaster

Real documented incidents. A coding agent deleted production data. Others followed the same pattern. The models worked fine. The systems around them did not.

Read on LinkedIn open_in_new
05

From Vibe Coding to Governed Systems

What it actually takes to move beyond vibe coding into development that holds up under real operational pressure.

Read on LinkedIn open_in_new
04

The Rise of AI Fluency and the Risk of Cognitive Drift

AI fluency is rising. So is the risk of offloading thinking that should stay human.

Read on LinkedIn open_in_new
03

When AI Moves Into Our Devices

AI is moving out of the cloud and into everyday devices. What changes in practice and what it means for how people learn inside organisations.

Read on LinkedIn open_in_new
02

When AI Stops Being Invisible

AI systems behave unpredictably more often than most teams expect. Sometimes invisibility is exactly the problem.

Read on LinkedIn open_in_new
01

From Prompts to Context

The first edition. A shift from prompts to context, a Make prototype and one question worth sitting with.

Read on LinkedIn open_in_new

This newsletter is published in English only.