Tooling

4 Jun 2026, BST · 12 min read

Running Local AI Models on macOS

Install, configure, and run local AI on Apple Silicon, with memory-optimised settings and VS Code Copilot integration

How I set up Ollama on an Apple Silicon Mac, tuned the environment variables for 16 GB unified memory, chose a model set for daily use and coding, and wired it all into VS Code Copilot using capped context variants.

2 Jun 2026, BST · 11 min read

GitHub Copilot for Engineers: Getting Better Results

From first install to a setup that scales: models, instructions, agents, and MCP

A practical guide to configuring GitHub Copilot effectively: choosing the right model per task, layering global and project rules, building a custom instructions and agents setup, and using MCP servers without accumulating noise.

22 Apr 2026, BST · 17 min read

Playwright E2E testing AI skills: JavaScript London talk

Using Playwright and the playwright-explore-website skill to test real user journeys

Companion post for my JavaScript London talk at NewDay on 29 April 2026, covering how Playwright and the playwright-explore-website skill support E2E testing, exploratory QA, and browser debugging.

22 Apr 2026, BST · 6 min read

The playwright-explore-website Copilot skill

From the awesome-copilot example to a stricter Playwright MCP workflow

How I set up a personal playwright-explore-website GitHub Copilot skill, starting from the awesome-copilot example and adding MCP bootstrap instructions, stricter rendered-UI review rules, safer browser exploration, and cleaner Playwright session outputs.

31 Mar 2026, BST · 3 min read

Keeping your zsh config out of Copilot's terminals

How a single env var separates your shell from the agent's

GitHub Copilot agent terminals are non-interactive and don't need your Powerlevel10k prompt or aliases — but a naive guard breaks your regular VS Code terminal too. Here's the fix: a custom env var that only VS Code injects into user-opened terminals.