The augmented founder has become the norm. The augmented team is still the exception. And that gap is exactly where the gains evaporate.
Guest post: Pierre Beunardeau runs Origin Education, a Qualiopi-certified AI training organisation, part of an ecosystem that partners with Ember.
Solo leverage does not transfer by osmosis
If you use a tool like Ember, you already know the leverage: a structured business plan in hours instead of weeks, a deck that holds up, qualified leads without an army of interns. An AI-equipped founder now does, alone, the work of a small team from five years ago.
Then the company grows. Three hires, then eight. And something strange happens: productivity per person drops. The founder notices the new hires use AI "a bit", "each their own way", "when they think of it". The personal edge did not transfer. It diluted.
National figures confirm the intuition. According to a Microsoft France study published in February 2026, 61% of employees who use AI at work do so through personal accounts, outside any framework, and 71% of non-executive managers have never received AI training. Your people already use these tools, but without method, without data rules, and without reproducible gains.
Why "they can just watch me" fails
Your practice is unreadable from the outside. Months of iteration produced your prompting reflexes, your guardrails, your nose for a doubtful answer. What the team sees is the output, not the method.
Every role has its own cases. Your founder shortcuts say nothing to your sales rep about lead pre-qualification, or to your developer about assisted code review. Without a per-role translation, everyone reinvents, badly.
Nobody wrote the rules of the game. Which customer data may enter which tool? What gets verified before it ships? Without explicit answers, every employee improvises their own security policy. That is shadow AI: invisible until the incident.
What works: treating the skill as an asset
1. Data rules before tools. One page is enough: what is allowed, what needs sign-off, what is forbidden. The company AI charter template published by Origin Education (in French) gives a directly adaptable base. Counter-intuitively, the frame unlocks usage: people dare to use AI once they know where the limits are.
2. Per-role training on real cases. Not a generic demo: the sales rep works on her follow-ups, ops on their procedures, the developer on the codebase. That is what a serious in-company AI training looks like: every participant leaves with skills they apply the next morning, on their own job.
3. Measurement at 30 and 90 days. A training that has not changed how the work is produced a month later was a presentation. Simple indicators suffice: time saved on the targeted tasks, weekly usage rate, quality as judged by peers.
The French bonus: the state co-funds this upskilling
For the existing team, a funding request can be filed with your OPCO under the skills development plan; the decision and amount depend on your branch and your file. And for a future hire, France Travail's POEI scheme can fund up to 450 hours of training before the employment contract starts, AI skills included: the details are in this guide to POEI applied to AI (in French).
A startup that hires and trains at the same time can have a real share of the effort carried by public schemes, provided the files are set up before signing.
The question to ask yourself this week
If you left your company for a month, how much of your AI leverage would keep working without you?
If the answer is "almost none", your advantage is a personal talent, not a company asset. The conversion is a bounded project: written rules, per-role training on real cases, a 30-day measurement. Founders who have done it do not go back: the whole team starts moving at the speed they thought was theirs alone.
Pierre Beunardeau is the founder of Origin Labs, a French studio building applied AI products and an Ember ecosystem partner, and of Origin Education, a Qualiopi-certified training organisation that trains company teams across France on their real business cases.
FAQ
Why don't a founder's AI gains spread to the team?
Because the practice is unreadable from the outside: months of iteration produced invisible reflexes. Without a per-role translation and explicit data rules, every employee reinvents, badly, on their own.
Where should a team's AI training start?
One page of data rules (allowed, needs sign-off, forbidden), then per-role training built on each job's real cases, and a usage measurement at 30 days. In that order: the frame unlocks usage.
Can a French startup get this training funded?
Yes, two main routes: an OPCO funding request for the existing team, depending on branch and file, and France Travail's POEI scheme to train a future hire before the contract starts, up to 450 hours.