My business, built alone with AI agents

In 2023 I started my own training and digital-product business. At the end of 2023 I left Avanade, after five years helping large clients’ teams adopt new tools. There was no IT department any more, no developers, no PMO. Product, sales, suppliers and budget all rested on one person.

The team that didn’t exist

A small business needs what a project team usually provides: a website, a catalogue, payments, booking, forms, automations. I built all of it alone, with AI coding agents and standard services: Cloudflare for hosting, Stripe and GoCardless for payments, Cal.com for booking, Tally for forms, Make and Zoho for automations.

The site you are reading was built the same way: close to 50 pages in French and English, a Lighthouse score of 100 on mobile, no technical team.

What Avanade had taught me

AI did not spare me the project manager’s work. It made that work essential. At Enedis, as proxy Product Owner on an application with 2,500 users, I wrote user stories and acceptance criteria, and every release went through user acceptance testing before going live.

With a coding agent, the discipline is the same. I describe the expected result, and how to check it, before the first prompt. I cut the work into small deliveries. I test each one before moving to the next. An agent quickly produces something plausible; without written criteria, there is no way to tell whether it is right. I describe the method in a piece on defining "done".

Train, then check

In the same business I trained close to 40 professionals one to one, and created a paid online course for a UK education platform. It confirmed what I had seen on client engagements. People learn a tool by using it on their own work, not by watching a demo.

What it means for your teams

I use AI every day to deliver, not to experiment. With teams getting Copilot or ChatGPT, I start from their real tasks, define with them what a good result looks like, then measure whether the tool helps.

I recently read 146 French job ads for AI adoption roles. 63 ask the hire to train users; only 17 say how usage will be measured. Those 17 are the ones that interest me, and that is the work I offer through ALGS services.