2016
Started working closely with the newly-formed data-science team at the Royal Society of Chemistry. Back when "AI" still meant machine learning and the tooling was rough.
About
Most AI advice comes from people selling their own product.
I'm James Stevens. For 27 years I've been the person inside businesses working out what tech is worth using and what isn't.
16 years optimising content platforms. 10 years as a product manager. Two years training product people and another two leading them. One thread through all of it: working out what to build, what to buy, and what to AI.
What technology belongs in a business — and what doesn't. That question has followed me since 1999, when I was an IT operations manager rolling out software for British Airways. It runs through a web agency of my own, 10 years as a product manager, two years teaching product management to publishers as a course tutor, and — currently — leading product at Time Out.
Along the way I've sat in most of the seats around the AI question: the person who has to use it, the person who has to buy it, the person who has to answer for it to the board. Commercial and non-profit. Engineers and data scientists. CEOs as stakeholders.
Admin Agents is how I bring that experience to managers and directors who don't have time to do the homework themselves. I've been building with AI for long enough to know what actually works and what's hype — and honest enough to say which is which.
Why I started this
Whatever your position on AI, one guarantee: someone in your team is already using it.
The copywriter who pushed AI-written content to a site that sells itself on human journalism. The PM whose bug report cites a tech stack that doesn't exist. The developer building with AI at home, while hand-coding at work, because Claude isn't approved. Whispers of "AI slop" in the corridors. The leadership too — AI-generated performance appraisals. Generic ChatGPT advice to professionals with marketing qualifications.
Good judgement about "When to AI" is a critical management skill.
The advice out there comes from two camps: vendors selling a product, or consultants selling a six-figure transformation. Neither helps the marketing manager whose CEO just said "use AI," or the director with a board paper due Friday and one analyst stretched too thin to help. That's who Admin Agents is here for — people who want to unpick what's already happening in their team and work out if it should be done, how to do it better, make it safer, without it taking a whole week to research and define it.
Battle scars
A decade working with AI — before it was called AI.
Not everything on my CV is relevant here. These are the moments that shaped how I think about AI, governance, and what to trust.
2018
First production ML integrations in editorial workflows — licence and article payment systems rolled out across 42 journals.
Nov 2022
Signed up to ChatGPT in launch week. Paid subscriber from April 2023 onwards. Built it into my workflow long before it became a household name.
2024
Built a framework for spot-checking ML decisions inside automated workflows — before "AI governance" was a phrase anyone used.
Mid 2024
Built a spelling tutor site for a primary school, and an automated golf-sweepstake tracker for a friend. Both taught me more about LLM limits than any whitepaper.
Feb 2025
Initiated Time Out's Google AI programme — worked on content, translations, integrations and operational cost reduction across global markets.
Mid 2025
Delivered a 45-point NPS improvement through customer experience transformation across 14 global markets — unified critical journeys from media platform to loyalty programmes.
Late 2025
Shipped global GDPR / privacy rollout across all EU traffic in two sprints — 25% revenue uplift and improved legal compliance. Introduced the engineering team to Claude.
Mar 2026
Rebuilt the sweepstake tracker as a production site over a weekend with Claude Code. The moment "prompts" became "agents" in my head.
May 2026
Briefed C-suite on ways to improve margin through AI data analysis and agent deployments in teams. Lobbied for AI tool rollouts outside of engineering and coached teams on AI best practice.
Apr 2026
Admin Agents born.
Next step
If any of that sounds like what you need —
Start with a 30-minute AI Primer. No prep required, no sales pitch, no six-figure proposal at the end.