01
Start with the work
Which task is changing, who does it today, and what would a better result look like? Write down the current process before selecting a tool.
2026 Briefing
What I’d tell your board
if they asked.
Four questions to settle before committing to an AI programme.
2026 Reality Check
02
Decide what the system may do
Define which actions an agent can take, which need approval, and how a person can stop or reverse them.
03
Inspect the vendor’s evidence
Ask for a working example in your environment. Check the cost, failure cases, data handling, and who will maintain it after launch.
04
Set a stop condition
Agree on a measurable result and a review date. If the pilot misses the mark, you should know whether to change course or stop.
A useful pilot produces a decision, even when the decision is to stop.
Selected Writing
Jan 30, 2026
I Build AI Systems. My Son Is Three.
On building AI systems while raising a child who will grow up with them.
Read on Substack →
Dec 16, 2025
I built a custom CRM last weekend.
What I learned building a CRM over a weekend.
Read on Substack →
Nov 17, 2025
The Missing Playbook for Agentic Systems
A practical framework for deciding where agents belong in a product.
Read on Substack →Get In Touch
Book a 20-minute call with me.
Tell me what you are working on. I’ll say where I can help.
Or email directly: me@jameshaliburton.com
Not ready? Subscribe to my writing or follow on LinkedIn