The one-man band: rebuilding Vouchernaut in 22 days
How I rebuilt Vouchernaut, the UK voucher site I started in 2016, as fifteen live products in 22 days, doing the product, design, engineering and operations myself while AI agents did the typing.
Notes from the work
Writing about client projects, product development and growth. The articles include project accounts, technical guides and notes on working with agents.
Featured article
How I rebuilt Vouchernaut, the UK voucher site I started in 2016, as fifteen live products in 22 days, doing the product, design, engineering and operations myself while AI agents did the typing.
12 articles
How I rebuilt Vouchernaut, the UK voucher site I started in 2016, as fifteen live products in 22 days, doing the product, design, engineering and operations myself while AI agents did the typing.
How I built Tau Ceti IV, a three.js map explorer for Bungie's Marathon with a Suno soundtrack, in three days with Claude Code, and why taste can lead a project now, not just technical ability.
How I rebuilt Benson, the coupon-extension blocker I first launched in 2022, as a full SaaS platform with a team of AI agents: 44 specs, 836 commits and a launch campaign in under a fortnight.
How I took Dinners With Friends from a napkin sketch to a booking business, then built the payments, operations and search programme behind it.
Researching 162 companies with Claude Code agents: a swipe tool instead of listing-site filters, an A to F reputation score instead of Glassdoor, and a fifty-line Puppeteer service instead of a screenshot API.
How to audit, secure and ship an AI-generated codebase: the five-minute Supabase check, what a proper review covers, who to hire and how long it takes.
An open-source Puppeteer wrapper that adds curved mouse paths, uneven typing and idle jitter, so legitimate automation stops looking like a bot.
A custom Chrome DevTools extension cut GrowthRunner's analytics auditing time by 90%, taking each specialist from one audit a day to ten.
How a Chrome extension built for Contra turned more than 35,000 users' own LinkedIn and X sessions into a discovery network that found more than 5 million opportunities.
What really happens when you take a Lovable prototype to production: three months, over a thousand users and the hard lessons learned along the way.
How I set agents and automations up inside a team: a catalogue of small reusable building blocks, the four levels of working with automation and how to measure whether it pays.
A one-week proof of concept that watches YouTube for a learning platform's content team: Trigger.dev queues, structured AI analysis and a 200x faster research loop.