Writing

Most conversations about AI at work are about delegation: what to hand off and what to keep. But the real risk isn't bad code, it's slowly losing the mental model of your own system. By planning in detail, following the implementation and stepping in when needed, AI becomes a collaborator that stress-tests your thinking instead of replacing it.

We agree on the psychology: Miller's 7±2, Hick's choice penalty, Gestalt proximity. Yet production apps rot into cognitive sludge. The friction isn't ignorance; it's that static psychological laws meet an append-only time series of feature requests.

Most developers use AI coding assistants as high-speed autocomplete to generate repetitive unit tests. But unit testing only validates what you already expected. By shifting AI into an adversarial 'red team' partner, we can generate property-based tests that uncover hidden edge cases, invalid state transitions, and boundary conditions before they reach production.

Software architecture is the clinical process of choosing which specific set of problems to manage rather than searching for a perfect solution. This approach prioritizes clear boundaries and system legibility to ensure that the internal structure remains a maintainable asset instead of a technical liability.

Shifting the accessibility narrative from a compliance chore to a high-level engineering constraint, exploring how color blindness and screen reader logic force better information architecture.

We’ve inherited the "handover" from the world of physical manufacturing. In that world, a blueprint is finalised, signed off, and sent to a factory floor where the variables are controlled and the outcome is static. But software is not a toaster; it is a living system of state, fluctuating data, and unpredictable human behavior.