Key Insights from XAnge’s AI Breakfast among peers – July 2, 2025
As part of our ongoing support to our portfolio companies, we hosted a breakfast focused on a challenge for many: “How to build an AI roadmap when your startup isn't AI-native?”
Participants:
- Baptiste, Founder & CTPO, Reflect
- Bastien, Founder & CTPO, Ouihelp
- Adrien, Head of AI Projects, 360Learning
- Jerome, European Startups Partnership Manager, Stripe
🔍 Quick access:
- 🎯 Why startups are integrating AI
- 360 Learning: strategic differentiation & competitive pressure
- Reflect: product vision & experience enhancement
- Ouihelp: unlocking previously inaccessible opportunities
- Stripe: internal enablement
- 🧩 Augmenting the right problems with AI
- 🛠 Building without a dedicated AI team
- 🚧 Challenges and cautionary tales
- 🔮 Next Steps
- 🧰 Complementary Resources
🎯 Why startups are integrating AI
Diverse, yet complementary motivations for integrating AI into product roadmaps:
360 Learning: strategic differentiation & competitive pressure
Reflect: product vision & experience enhancement
Ouihelp: unlocking previously inaccessible opportunities
Stripe: internal enablement
🧩 Augmenting the right problems with AI
Participants emphasized a pragmatic approach to deciding what deserves to be "AI-ified":
- Criteria for Prioritization:
- Frameworks for Sanity:
🛠 Building without a dedicated AI team
A recurring theme was scrappiness over scale:
- Team re-skilling over hiring: try to favour empowering existing team members to explore AI with available tools instead of hiring data scientists.
- Avoid illusions of productivity: superficial productivity gains from AI can lead to cognitive atrophy, as teams stop deeply understanding their tasks and tools and become over-reliant on outputs they can’t debug.
- Team members buy-in and adoption: use creative methods like AI hackathons, usage incentives, and internal icebreaker agents were used to engage reluctant team members and demystify AI.
Ex: Using AI to replace traditional analytics queries—freeing up data teams for more critical work.
One founder highlighted the “lightspeed tech debt risk”: delivering fast but accumulating long-term maintenance challenges.
🚧 Challenges and cautionary tales
Several common pitfalls emerged from collective experience:
- Platform legacy constraints can make it hard to iterate quickly, especially with low-quality data
- User exposure to AI features w/ validation can lead to poor performance, moderation issues.
- False sense of autonomy as over-reliance on AI can lead to decisions that are hard to trace or understand. You still need to be able to develop critical thinking.
🔮 Next Steps
Participants shared that they’re still exploring and questioning themselves about:
- How to scale AI safely across their orgs without losing product quality.
- How to ensure maintainability and documentation as more code is generated / integrated by AI.
- How to rely on open-source frameworks to avoid lock-in with proprietary tools.
🧰 Complementary Resources
The Interview with Arielle Le Bail, Product Lead France at Stripe to deep dive on How Stripe Builds with AI ?
Tools we talked about during our breakfast
Tool / Framework | Use Case |
Dust | Widely used but facing constraints as major vendors close off APIs to push native solutions |
n8n + Zendesk | Automating agent flows for support |
Qube | Semantic layer abstraction to better prompt LLMs |
Gong | High-quality multi-level meeting summarization, plus “devil’s advocate” feedback loop |
SuperWhisper / Oyama (local) | Meeting recording with no data leakage |
The Siparex Operating Team Guide around GenAI. Tools, use-cases, deep dives: everything you need to start implementing AI, no matter your knowledge on the topic.
Tech.eu podcast’s with Laura Modiano from Open AI. If you want to deep dive on what EU Startups do on AI.
