B-01
Rules before reasoning.
If something can be written as a rule, write the rule. I keep seeing LLMs given jobs that a few lines of logic would do more cheaply and get right every time. Save the model for the parts that need judgement.
01 / Hello
AI models are rarely the hard part.
Hi, I'm Nick. I lead AI delivery in complex settings. At work I'm the one pushing everyone to try AI on their real jobs, and also the one explaining why a chatbot isn't the answer to everything.
- NS-001 / Day job
- Senior Manager, Capital Markets Technology, EY
- Before
- Royal Bank of Scotland. Before that, a captain in the Irish Guards.
- Builds
- Agents, mostly. Then the agents build the rest.
- Asks
- "Could an agent do this?" Then, "Would a rule do?"
02 / Who
I started building for the web as a teenager in the late 1990s. Dreamweaver first, then Flash and ActionScript, then back ends from scratch and every kind of CMS going.
Then came the Army, a bank, and consulting to banks, where the job is getting complicated change to land somewhere that can't afford to get it wrong.
Now agents write most of my code. Teenage me would have loved that.
Fig. 02 / Changelog
Built with Dreamweaver, then PowerPoint, and now agents.
03 / Beliefs
Things I believe
About AI, about leading people, and the places where the two overlap.
B-02
A longer prompt isn't a harness.
When an AI tool gets something wrong, the reflex is to add another paragraph to the prompt. What it usually needs is structure around it: checks before and after, a clear hand-off, and a person signing off anything that matters.
B-03
Tell people why. Let them work out how.
Mission command, which I took from the Army: give people the intent and the limits, then get out of the way. Most managers give too much how and not enough why. It works on agents too.
B-04
Agents can write the code. I still own what ships.
A lot of the code on my GitHub was written by agents I run, and the commit history says so. I set the spec, approve each phase and read what comes back. If it breaks, it's my name on it.
B-05
Calm on top. Paddling underneath.
People who've worked with me tend to describe me as calm. That's the top half of the swan. The Army taught me that a room copies the mood of whoever's in charge, so the paddling stays out of sight.
04 / Built
Things I've built
Agents wrote most of this code, and the commit history says so. What I built is the system around them: the specs, the approval gates and the review. I still own what ships.
The Foundry
Featured build
An overnight software factory. A cheap model reads developer complaints on Hacker News and Reddit, a stronger one picks a problem and writes the spec, and a coding agent builds an MVP and pushes it before I'm up. I commissioned it, approved each phase and reviewed what it shipped.
01 Scout
Claude Haiku
Reads the complaints
02 Spec
Claude Sonnet
Picks one, writes acceptance criteria
03 Build
aider + Sonnet
Writes the MVP
04 Ship
git push
Before breakfast
Day job, approved
The work one
An agent system for the admin layer of a consulting engagement. It turns meeting transcripts into actions and drafts reports, plans before it acts, waits for my approval, and cleans anything before it leaves the firm.
05 / Off the clock
Off the clock
- Three kids, one Labrador, and a lot of walks.
- Painting Warhammer miniatures. You'd think the Army would have cured me of moving small armies about. It didn't.
06 / Now
Updated September 2026
Now
- Setting up Hermes agents to take on the home admin.
- Working out my own operating model with AI in it, for work and for home.
- Iterating this site until it feels more like me than a CV.
07 / Say hello
Say hello.
If you want to talk about AI delivery, agent systems or leading teams, email me.