What we've learned, written down.
We run our own business on this and we test other people's. These are the things worth writing down: what works, what quietly doesn't and what isn't worth doing at all.
The spreadsheet that quietly runs your business
Every established business has one. It works, it's critical and one person understands it. How to tell when it stopped being a spreadsheet.
The work that isn't worth automating
The shortlist everyone wants is what to automate. The more valuable one is what to leave alone. There are four reliable tests for telling them apart.
An AI use policy people will actually follow
Most AI policies are written to be defensible rather than useful, so staff route around them. A shorter one answering the real questions works better.
What actually goes wrong in AI-written code
Not exotic failures. The same handful of gaps turn up again and again, they all look reasonable in review and they share one cause worth understanding.
Shadow AI is already in your business
Your staff started using AI months ago. The question is whether you find out what they're pasting into it before somebody else does.
Why automation projects stall at eighty per cent
The build works. The pilot went well. Then it sits, nearly finished, for four months. The cause is almost always the same and it isn't technical.
Prompt injection, in plain English
If your AI tool reads anything a stranger can write to, then a stranger can give it instructions. The shape of the problem, without the jargon.
Buy, build or neither
AI made building cheap enough that the old answer stopped being automatic. It also made the third option, changing nothing, easier to overlook.
Got a question none of these answer?
That's usually the more interesting conversation. Tell us what you're weighing up.