AI exploration and working in the open

I'm collecting stories of how people are using AI in enterprise setting. These can be few and far between because many companies don't work in the open despite the benefits of working in the open.

October 2025

Love Holidays

George Malamidis has been journalling Love Holiday's journey towards an AI first company:

• https://www.linkedin.com/pulse/loveholidays-ai-diary-0-george-malamidis-2qbpe/
• https://www.linkedin.com/pulse/loveholidays-ai-diary-1-george-malamidis-csnie/
• https://www.linkedin.com/pulse/loveholidays-ai-diary-2-george-malamidis-xm3ae/

I like this important principle:

We're not doing AI for the sake of AI

Before diving in, it's worth highlighting this core principle. Every AI initiative at loveholidays must improve customer experience and/or business value. For engineering, it must also measurably enhance developer experience and efficiency.

NHS England

Tom Hallam wrote a month-note (like a weeknote but capturing a month) about AI usage at NHS England, with a deep-dive into the User Research Finder and Copilot projects.

He knows what he wants to do but you can't simply vibe code your way to a solution. His reality is using the tooling that is available to him. I shuddered at the mention of Sharepoint! He has some general advice, including:

Tips for building products or tools that integrate with AI/LLMs

• Be Agile: Reduce your scope, the MVP you can actually build, maybe not the Most Useful Thing for your end users. As your team learn more, and get more buy in, then increase the scope.
• Listen: Do regular research. Share early and often with your team and stakeholders. Getting feedback is critical for improving the service.
• Data Data: Think like a data scientist and work like a data engineer:
  • Plan out the system architecture and data flows
  • Build in the information governance processes
  • Consider security - what type of data, where is data stored, who has access to it
  • Use some version control and create backups
  • Test and learn, avoid making too many changes at once.
  • Test your prompts with lots of different inputs.
  • Create a test pack covering happy paths and edge cases. Check things work every day.
  • Expect things to break, have a recovery plan.
• Have fun - while this is a challenging space to work in, when the solution works it is really rewarding to hear positive feedback.