Chat Transcripts, AI, and BDD: Escaping the Feature Factory Trap
When I look back at some of the organisations I have worked in, especially larger ones in financial services, one pattern really stands out: the feature factory. You probably know the setup. Product is a silo. Engineering is a silo. QA is a silo. Work is passed down the line...
How to use meeting transcripts, AI, and Behaviour Driven Development to transform product development from a feature factory into collaborative, value-driven engineering.
Case Study: Building a Fully Self Hosted, GDPR Compliant RAG Pipeline
When I started working with this client group, the brief was straightforward. They wanted to see if historic support ticket data could be used to help their staff resolve new tickets more quickly. There were six years of history sitting in their systems, which meant a huge body of knowledge that was not being put to use.
There was one major catch. The tickets were full of personal and sensitive information. Under GDPR, sending any of that data to OpenAI or another cloud provider was a non-starter...
Back to Cursor: Testing GPT-5 for Coding
A few weeks ago, I received an email from the Cursor team that made me stop what I was doing. For a short window of three or four days, GPT-5 would be free to use inside Cursor. I had already been curious about how GPT-5 would perform as a coding assistant, so this was the perfect opportunity to test it.
I wanted to see whether GPT-5 could work effectively inside a manually structured project, and whether the experience would be noticeably different from Kiro's Claude Sonnet 4 preview mode. The results were interesting...
Kiro: Amazon's Quietly Brilliant Entry into AI-Assisted Coding
As someone who regularly experiments with new AI tooling, I was curious to see how Kiro compares to more established options. I'd been a long-time Cursor user - drawn to its deeper AI integration and workflow support - but recent pricing changes made me reconsider.
Kiro had just launched, it was free to use during preview, and it promised a more structured, spec-driven development experience powered by Claude Sonnet 4.0 / 3.7. That was more than enough reason to take it for a spin...