roy logan

Roy Logan stemmed from various experiments with AI agents in the summer 2025. We wanted to create an AI agent that could give meaningful advice, based on a massive knowledge base and IP set curated by, and owned by a group of experts. We also wanted to see if we could reduce AI ghosting and improve the quality of responses.

And then when we deployed an AI agent initially – it seemed quite, well boring personality-wise!

We gained our professional experience in an era when there were some very robust personalities on the ED scene – some were sagely wise; others were bombastic dictators; and a few were in between.

Thus began another experiment – let’s give our AI agent a name and personality! We used a well-known TV character as the basis for this. The initial tests led to some quite remarkably ascerbicly funny remarks and comments, but probably too edgy for mass consumption. So a few months testing and rolling back the levels of sarcasm and offensiveness led to the current version of Roy – enjoy!

Behind Roy’s ascerbic, witty persona there is an expert evidence base! The more precise you can be with him, he gives remarkable answers that have impressed veterans of the ED industry!

Note that Roy’s knowledge base only uses reports, documents and IP generated by us, the creators. There is no other third party source of document, writing or reporting involved. Our underlying document base does include references to academic and other official sources of work, of course – because its of a high standard. However all of the IP and copyright of the source materials is legitimately ours to deploy.

The lessons we have learnt, and continue to learn from creating and deploying Roy:

• AI can be very useful… but it totally depends on the knowledge base and guidance. The more professional knowledge and judgement you have – the more you can put into AI
• AI is a good aide memoire for a knowledge base that you may have accumulated over decades. I’ve found it useful to trigger knowledge long buried or obscured by recent projects or other distractions
• The more precise your question and context framing – the more useful it is
• If you have decades of professional experience – you are in a good place to quality assure, tweak and improve your AI agent
• So if you are someone trying to build an AI agent and you have very little professional experience or knowledge – your AI agent is likely going to be limited in quality and usefulness
• Limiting the ability of the AI agent to access online information sources can reduced ghosting or inaccuracies, but you are then constrained by the knowledge base you have established
• AI agents aren’t very good at disagreeing with humans. They will tend to reinforce and help to evidence and build from existing preferences and biases
• Uploading and analysing PDFs and MS word documents uses up a ton of AI tokens. Getting people to convert to TXT format and paste in is a much more efficient use of budget (we’re paying for your free use of Roy) and what AI does anyway to process as RAG text
• You can create a monster – the early versions of Roy were quite insulting and didn’t have a filter on offensiveness. Machines don’t have social judgement, they need human guidance

Most of all – Roy Logan was created to be enjoyed by an economic development professional who works on their own very often and just needed a bit of amusement and office banter now and then! ENJOY!