$ tail -f blog.log▋
Field notes on AI that actually works
Trends, problems, certifications, and the occasional opinionated take, written from client work, not from a content calendar.
- Responsible AI10 min read
TypeSafe's Jev is a model that doesn't write. Here's why that matters.
TypeSafe AI opened early access to Jev, a decision model that returns probabilities instead of text, in September 2026. Research on inference energy suggests models like this could cut the cost of a lot of everyday automation.
- Automation in practice8 min read
Is this task worth automating? What the field studies say
Three large studies of generative AI at work point to the same pattern: the gains are real, uneven, and depend on the task. Here's what that means for choosing what to automate.
- AI education6 min read
Should you get better at prompting, or better at judging the output?
Prompting tricks go stale with each model release. The research on automation bias and AI-assisted work suggests the skill that lasts is knowing when an answer is wrong.
- Responsible AI7 min read
What AI governance actually looks like outside a Fortune 500
NIST, ISO and the EU AI Act all assume a compliance function. Strip them back and they ask three questions any business can answer. A court case shows why the third one matters.
- Problems with AI6 min read
The context window trap: why a bigger model doesn't fix bad retrieval
Long-context benchmarks show models losing track of facts buried in the middle of their input. Pasting the whole knowledge base into the prompt walks straight into that.
- AI trends6 min read
Agentic AI: what the term actually means, and what it doesn't
Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027, and says most vendors using the word aren't selling agents at all. A look at where the definition landed.
- AI trends5 min read
How to actually keep up with AI (without the doom-scrolling)
The biggest AI story of the last few years for most businesses wasn't a benchmark. It was a roughly 280-fold fall in what a given level of capability costs to run.
- Responsible AI8 min read
The environmental cost of AI, and where it actually sits
Data centres used about 1.5% of the world's electricity in 2024, and the IEA expects that to roughly double by 2030. The research says the part most businesses control is inference, and model choice.
- AI trends6 min read
What's production-ready in AI automation right now
A mid-2026 status report: what has become routine engineering, what is still fragile, and what the analysts expect to fail.
- Problems with AI7 min read
Where AI automation breaks (and how to catch it first)
A decade of research on machine learning in production says the model is the small part. The failures live in the data, the testing and the ownership around it.
- AI certification6 min read
AI certifications in 2026: what's changed, and what's worth your time
AWS retired its Machine Learning Specialty, Microsoft replaced its Azure AI Engineer exam with one built around agents, and the EU's AI literacy rule is driving demand for basics. A status update.
- AI education7 min read
AI automation in education: where it fits and where regulators draw the line
UNESCO's guidance, the UK Children's Code and the EU AI Act all point the same way: AI can take on school admin, but decisions about students need a much higher bar.