2026: agents that run, not slides.
I design, code and put into production, myself, agents for what comes back every day in a company: customer requests, content and creative work, brand documents, checks, visibility inside AI answers. Twenty years of premium brands behind, a proof log in front.
Hello, we are looking for a yacht for eight people, 12 to 19 July, leaving from Ajaccio.
Twenty years leading premium brands in agencies: Cartier, Nespresso, Jaeger-LeCoultre, Clarins, L'Oréal, Hermès, LVMH, Coca-Cola, Heineken, SEB. Since 2019, systems built with my own hands and in service: Corsica Yacht Charter, Ilka Schumacher, GeoScope, ACT, eu-verify.
Every week brings a new demo, a new tool, a new promise. Meanwhile, customer requests still wait for an answer, content is still made by hand, and nobody knows what AI assistants say about the brand.
The problem is not choosing a tool. It is getting something useful to run, safely, with someone accountable for what goes out.
A demo is not a production system.
An agent without human validation eventually sends something stupid to a customer.
AI on top of broken processes creates nothing clean.
A brand absent from AI answers gets replaced by a more visible one.
Before talking about AI, I look at what your team really does in a day, whatever the job. Not the job title: the tasks, one by one.
Calling a loyal customer, deciding on a goodwill gesture, settling a delicate case. This is where your team makes the difference, and the agent stays out of it.
Sorting, qualifying, checking, drafting, summarising. The agent does the preparation work, at any hour, the same way every time.
Everything that leaves the company passes in front of someone on your team. They reread, correct if needed, and they are the one who sends.
What the agent prepared, who validated, what went out: everything is recorded in the proof log. It is what tells you whether the pilot delivers.
Eight days to read your real flows, customer requests, content, tools, data, and choose the automation that pays the most. Then an agent in production in ninety days, with the expected result written down before we start.
Three to four days a month, reporting to the owner: I build, I train the team, I hold the governance. Code and documentation stay with you.
For an agency or a marketing team in Paris: three days a week on site, one agent billable to a client within ninety days, then creatives and developers who can run it without me.
Three days per marketing team to understand what an agent does, what it must never do, and how it is controlled. Useful for the AI Act training obligation.
It handles a real flow: inbound requests, content, checks. Every action is logged, every exception reaches a person.
Nothing goes to a customer, a website or a social network without a click from someone on your side. The draft is ready; the decision stays yours.
What was read, chosen, written, validated, sent: timestamped, reviewable, defensible. You always know why the agent did what it did.
Code and documentation with you, the team trained, a named technical relay for maintenance. You are not locked in with me.
Seven systems, seven different problems. For each one: the starting problem, what the agent does, and what proves it.
Built for the client
I do not turn a vague need into an 80-slide deck. I look at your real flows, pick an automation that pays, build it, put it into production with its guardrails, then we look at what changed.
Your inbound requests, your content, your tools, your data. What can be automated cleanly, what must stay human, what is missing.
One automation, one scope, one expected result written down before we start. Ninety days.
I code, I integrate your existing tools (e-mail, WhatsApp, website, payment, scheduling), I wire human validation and the proof log.
The agent handles the real flow. You validate. Every action is logged. We fix against reality, not in meetings.
Code and documentation with you, the team trained, a named technical relay. You can run it without me.
No demo sold as a production system.
No magic AI on top of a broken process.
No agent that sends anything without human validation.
No deliverable without a proof log.
No jargon to disguise a weak decision.
Twenty years of premium brands taught me what a word costs. Seven years of coding taught me what an agent really does. You get both in the same person, and that is rare.
Founder of GreenEris. AI studio director: AI agents and automations in production for premium brands, across marketing, brand and customer experience.
Twenty years leading brands in agencies (Nespresso, Cartier, Jaeger-LeCoultre, LVMH, Coca-Cola, Heineken, SEB, Clarins, L'Oréal, Hermès) as art director, international account director, partner, then brand experience director.
Since 2019, I build what I used to recommend: an AI lead-management system for Corsica Yacht Charter, a painter's website and studio copilot, GeoScope, ACT, eu-verify, Storyboard Director, and AAOS, my own standard for governing agent organisations. Nothing ships without human validation and proof.
Based in Ajaccio. Available remotely, as a fractional AI director, or in residence three days a week in Paris.
Owners of premium SMEs and mid-sized companies, yachting, hospitality, wine estates, cosmetics, prestige real estate, art, who receive requests, produce content and want an AI that works without putting the brand at risk. And agencies that sell AI to their clients and need someone who knows how to build it.
If you want a demo to reassure a board, this is the wrong brief. If you want an agent in production within ninety days, with its guardrails, we can move.
No. It prepares, proposes and logs. Someone on your side validates. That is the rule for everything I build.
Not to start: I build and integrate your existing tools. Afterwards, code and documentation stay with you, and a technical relay is named in the contract.
It stays with you. Access is limited to what the agent needs to read, every read and every send is logged, and nothing is used to train anything.
Yes, in the diagnostic: we choose what pays before coding. But I do not sell strategy without execution.
When the agent needs a place to work, yes. I do not sell a website, I put a system into production.
Remote by default, from Ajaccio. In residence three days a week in Paris for an agency or a team. On site for workshops.
Pick a flow, set three sliders. The answer is the one I would give you on the phone, only less polite.
Possible, to be checked in the diagnostic.
The potential is there, but the real flow needs a look: exact volume, tools in place, exceptions. That is what the diagnostic is for, and it may conclude this is not the right first topic.
If your customer requests wait, if your content is made by hand, if nobody knows what AI assistants say about your brand, the problem is fixable. Not with a demo. With an agent in production, validated by you, proven by its log.
