Lab Notes
Plain answers to real AI questions.
What we're learning about making AI stick, and plain answers you can use the next morning.
Latest notes
Why the same AI works better for one team than another
How relevant business context helps AI produce more useful answers, and how to keep it consistent across the tools you use.
The AI sounding board
From rough idea to worked-through plan in a day. How to use AI to get an idea into shape before anyone else sees it, why you have to make it argue with you, and what still needs your team.
Microsoft's leaky AI moat
Microsoft rarely wins with the best product. It wins by already being inside the business. Why Copilot's real advantage is your own context, and where that advantage is leaking.
Ten thousand lifetimes of reading
How AI models are actually built, explored in conversation with Claude. Why nobody had to label the internet, why the tools are confidently wrong, and why the bottleneck is you rather than the model.
I don't know where my work ends and the AI's begins
A presentation built in heavy collaboration with AI, and the unease on the way home. Where the line falls between the words and the decisions, and the four-part test I hold myself to now.
Does using AI undo your environmental commitments?
Your own AI use costs about thirty kettle boils a year. The industry's use is a different question, and it is being argued out in Scottish planning applications right now.
Stop hunting for prompt formulas. Start closing the loop.
Good AI work is not a straight line from prompt to output. It is a loop: brief it properly, demand honest criticism, get a second opinion on anything that matters, and keep going round until the work is genuinely yours.
How to teach Claude to write in your voice
Don't describe your voice to AI. Show it. Five steps, an afternoon of setup, and every draft after that starts from your voice instead of a stranger's.
How to write a prompt that actually works
The difference between a vague ask and a usable answer, with worked examples you can copy. Briefing well is the first step of a longer loop, not the whole method.
How to build a custom AI assistant without a developer
A working assistant, built in an afternoon, in tools you already have. No code required.
Microsoft Copilot for small businesses
What Copilot actually does inside Outlook, Word, Excel and Teams, what it costs, and how to decide whether it is worth it for your team.
What can Copilot do in Outlook, Word, Teams and Excel?
A tool-by-tool breakdown of where Copilot genuinely saves time, and where it still needs checking.
ChatGPT vs Claude for business
The two tools most teams shortlist, compared on the work businesses actually do: email, documents, analysis and research. The capability gap has closed, so the real question is which one your team will use well.
AI training in Scotland: five options compared
Free courses, public programmes, colleges, consultancies and private team training compared, with honest notes on what each is good for.
Find your starting point with AI.
Reflect on your own AI use with the free Personal AI Index. Organisations can also explore adoption and governance through Benchmark Lab’s indices. No training booking is required.