Things I Built That Actually Worked

I build a lot of things because I keep looking at perfectly ordinary business processes and thinking: surely we can make that less annoying.

Some ideas work first time. Most do not. This page collects the ones that survived contact with the real world.

The builds

• How I Built a Proper B2B Buying System for RKP
  Trade accounts, permissions, pricing, markets and the difference between an account existing and an account actually being ready to buy.
• Why I Built Shared Company Baskets for Trade Customers
  Because a £600 parts order should not disappear just because the person who started it shut their browser.
• How I Automated Paid Invoices Without Paying for Another Microsoft Service
  The obvious route wanted another paid connector, so I rebuilt the flow around tools we already had.
• How We Automated Paid Invoices From Shopify
  The other half of the invoice build: who gets the invoice, when it is sent, and how customers download their own copy.
• The Surprisingly Complicated Job of Chasing Unpaid Orders Automatically
  24 hours, 3 days, 7 days. Then we discovered that “not fully paid” and “unpaid” are not the same thing.
• Making Kart Parts Easier to Find When Customers Don’t Know the SKU
  Customers often know what a part looks like and where it fits. The website should understand that.
• I Built a Team of Local AI Agents to Check Each Other’s Work
  One does the job, one checks it, one reviews it. Then Windows refused to update the software because it was still running.
• I Accidentally Built a Go-Kart Game Inside an Ecommerce Website
  A 3D workshop, real kart parts, five marshals, a kart graveyard and a deliberately unreliable 10mm socket.
• How I Made a Browser-Based Go-Kart Look Worn Without Killing Performance
  More detail on the Kart Shed kart without the browser catching fire, and why a worn chain has to look more worn than it really is.
• How I Set Up Claude and ChatGPT So They Stop Agreeing With Me
  The personalisation framework I use to make AI challenge assumptions, separate facts from guesses and understand how I actually work.

The rule for this series

No glossy case studies where everything magically worked.

Each build log includes the bit that broke, the wrong assumption, the awkward workaround or the moment where the clever solution turned out to be more complicated than the problem.

Because that is usually where the useful lesson is.

Browse the full Build Log