Websites, an Android app, a smart home screen and two working agents, all built with Claude. I’ve captured how each works, the techniques that mattered, and what I would do differently next time.
Planning Defender, Section 20 Defender and School Place Appeal each turn a letter or notice (i.e. something that would usually require research and often legal expertise to respond to) into a structured, evidence-led pack. Claude built them from first idea to launch, and a second AI checked the content.
A custom launcher for four Fire TV sticks that opens my apps and runs the house from the remote. A Home Assistant bridge lets it control a vacuum that Homey Pro cannot.
Each morning an agent checks about eighty sources, reads every advert and scores it against my profile, then emails me the best with reasons.
An agent that reads my inbox each morning, keeps a full audit in a spreadsheet, works out who I need to follow up with and creates an initial draft of those emails for me. It also helps prepare me for the day's meetings by creating a pre-read ahead of each of my meetings.
Enter what a household earns, owns and spends, and see whether the money lasts. The report is built by plain code with no AI call, and a separate AI reviewed it.
A personal tracker for visiting Michelin-starred restaurants, filled by parallel research agents and checked by scripts. One source file builds both the web and the Android app.
A nine-page static site whose look comes from one rule: every image is generated from written brand guidelines.
The same projects, grouped by the approach that ran through them.
Choose a technique to see which projects used it.