Five ventures: AI Jungle Jungle Roots LearnAIThing Make At Home Go International Advisory
Loïc Guyon, international operator. Human judgment improves the agents. Their performance improves the business. Every cycle improves the operator.
An international operator. I build AI agents and workflows for businesses, and a named person keeps every decision that matters.
Two questions; the second depends on your first answer. The result names one place, quotes the two answers that decided it and gives a second link. Nothing is sent.
Taken from each venture's own site on 19 Oct 2026: its published entry price, and a dated, quantified result when it has published one. Where it has not, the line says so.
| Venture | Starts at | Published result | Source |
|---|---|---|---|
| AI Jungle | Leverage Assessment, from $999 | Asia-Connect, a named client: 755 executives contacted, 139 qualified meetings booked and 90 held in three months. Published 26 Jul 2026. | price result |
| Jungle Roots | Free first step, then the Shortlist Report at $499 excl. tax | No client result published. It publishes its own starting point: no mention of Jungle Roots in 13 usable AI answers, 1 Oct 2026. | price result |
| LearnAIThing | Builder Scan, from $4,900 | No client result published; the site says “Nothing here claims a client result.” | price result |
| Make At Home | Object design, from $90 | No client result published; the site says it has not yet tested the machines it presents. | price result |
| Go International Advisory | Market Intelligence Report, $3,450 fixed | No client case is public; the site publishes one only with the client's written approval. | price result |
Every venture runs the same loop. Each step is a rule I published, with the piece that backs it and one example from this site.
| Step | Without the clause | With the clause | Example on this site |
|---|---|---|---|
| DiagnoseProblem and baseline | Start without a baseline you can measure again. | Name the business problem and one observation you can repeat later. | Source watch: an agent watched public sources and isolated what changed, so specialists could interpret it. |
| TestTest set | Test only the easy cases. | Run normal, missing, ambiguous, contradictory and sensitive cases before anything ships. | The router on this page: every reachable path is checked against a written table of expected answers. |
| BuildChoose by the task | Rewrite what already works. | Assemble what already works. Do not rewrite it. | Escape experience: a booking assistant and an operating cockpit, assembled as a working proof of concept. |
| OperateHuman approval, failure and evidence | Let the system act with no named approver. | A named person approves, the stop route is known and the record can be inspected. | Network reactivation: the copilot drafted the first message; the owner chose who to contact and sent it. |
| ImproveOwner and review decision | Carry on without ever deciding to stop. | Each cycle ends on one decision: deploy, revise or stop. | 13 Sep 2026: the on-site guide chat had no conversation in 30 days. Decision: stop. It was removed. |
After Improve, the loop starts again at Diagnose.
Four accounts, anonymised. Each one states the role I held and the decision that stayed human. They are not independent validation of results.
It identified dormant relationships worth reopening and drafted the context. The owner chose who deserved contact and sent the final note.
It extracted prices, checked eligibility and assembled sourced technical notes. The team gained an audit trail.
It watched public sources and isolated what changed, so specialists could interpret the impact instead of discovering PDFs.
A themed customer journey, a booking assistant and an operating cockpit, built as a working proof of concept. It is not a delivered client site.
Errors, logged
Some names stay private. The scope of responsibility does not.See all selected work →
I am Loïc Guyon, an international operator. I work between countries and teams, between a technical possibility and a system people can actually run.
I publish the decision that changed, the error that cost time and the evidence that survived.