Émile Gadenne

Émile GadenneSep 18, 2026

We found the feature that's going to get us to product market fit. The problem with AI agents + internal data is rarely the tech. It's imagination. A business user sees what the agent can do. They get the idea. Then they open the chat. Blank field. Nothing. They don't know where to start. So they come back less. The tool gathers dust. Nobody understands why adoption stalls when the tech is right there. At Agrolytics, we had this exact problem with Marc, our sales agent. Clients knew he could spot a client slipping away. Prep a follow-up. Read the numbers. But writing the right question? That's a skill of its own. So we added a library of 150 prompts, right in the composer. One click. Real use cases. No blank page. And guess what ? Our clients love it And it taught us something: the bottleneck was never how powerful the agent is. It was the only a problem about education. Want to see what this looks like for your business? Calendly in bio ;)

Émile GadenneSep 25, 2026

It's been 9 months since I started building a bootstrapped startup, and here are the 5 things I wish I'd known earlier: 1. Customer feedback is gold. What will help you most is feedback from your ICP. But to get it, you have to convince them to try your product, or to buy it. Lack of credibility is often what holds startups back, even more so in high-ticket B2B. Nobody wants to bet on a product they don't know, built by a team they don't know. So if you can't sell yet, resign yourself to offering free trials. Even if it burns a bit of cash early on. Real user feedback beats ten hypotheses. 2. Prospect and talk to as many people in your industry as possible, every single day. Every conversation teaches you the vocabulary, the real problems, the priorities. Building in your corner something that isn't aligned with what's happening on the ground is wasted time. 3. Do more interviews early on. 100 minimum. It will save you time, and above all, pivots. Every extra interview is a bad assumption that dies before you've written a single line of code on top of it. 4. Don't hesitate to invoice. Being a startup is not a reason not to charge. Not invoicing means losing credibility. An invoice tells your customer your product has value. And it tells you whether that's true. 5. If you target a non-techy industry, go for outbound. Your customer isn't (always) on Product Hunt or scrolling LinkedIn all day. Phone and email are still the best way to reach them.

Émile GadenneSep 21, 2026

The AI in your ERP is already outdated. An AI feature announced (in your ERP) today was decided 18 months ago. In AI, that's several generations of models. I don't blame the vendors. I actually think they're doing exactly what they should. An ERP ships one major version a year, and every update is validated at the customer's site with production on the line. That's how you protect a system that must never go down. An ERP is built to last 10 to 20 years. An AI model lasts a few months. Two rhythms with nothing in common, and the gap widens with every lap. The business model points the same way. An ERP creates its value at deployment: you install it, configure it, then the system should move as little as possible, because every change costs the vendor and the customer alike. For me, useful AI is the opposite. It's not a feature you deliver, it's constant evolution. It tells you: this item is going to run short, this customer is drifting away, this margin is about to erode. And that can't be signed off once and for all. It's measured week after week, corrected, and retrained on what actually happened. Asking that of an organization built for stability means asking it to work against its own model. And I believe waiting has a real cost: customers lost without warning, margins given away on price increases spotted too late. That's why I think AI will only be mastered by startups. Correcting, retraining, changing course every week: for us, that's the job, not a constraint. And the ERPs that win the race, I'm convinced, will be the ones that team up with startups to adapt their products, rather than running alone. CETRA Informatique, the publisher of EURAGRO and our partner, made exactly that choice. Rather than building its own AI layer, it opened its ERP to us. And I think that decision is the reason they will gain market share. I've laid out the full reasoning in the article, link in the comments. Feel free to comment what you think about it!