Hi Reader,
Yesterday I was in Paris for the AI Now Summit, hosted by Mistral AI.
1,500 people, from industries and AI builders and start-ups (Neo4j, Snorkel AI, Microsoft, Qdrant, to name a few).
Genuinely interesting talks. Real use cases being shared.
Only one unforgivable oversight: not a single French croissant in sight! 🥲
I'm still processing that...
But, disappointment aside, here are my 4 key takeaways from the event.
1. AI is finally landing in conservative industries
I've been skeptical about AI adoption in highly regulated sectors. Honestly, still am, to some degree.
The gap between "demo" and "production" in industries like finance, healthcare, or energy is enormous.
But at this event, I heard about some concrete, real-world deployments that changed my view a little.
BNP Paribas is using AI to automate KYC (know-your-customer analysis for corporate clients).
EDF, the French state-owned nuclear operator, is working with Mistral AI to build tools that support nuclear engineering, maintenance, and reactor construction.
And Singapore's Ministry of Home Affairs has deployed a passport-less immigration clearance system at Changi Airport.
These aren't marginal productivity gains.
They're changes to how things actually get done.
Less bureaucracy. Less mind-numbing manual work. Less queing just because of the color of your passport 🤯. More time and resources for things that actually matter.
If you're tired of all the AI noise, the shiny object syndrome, the CEOs demanding "AI strategies" and forcing employees to reach AI token quota without any real ROI in sight, I get it. I'm tired of it too.
What actually changes lives is use cases like these. Not another chatbot wrapper.
And the ones that made it into production all had one thing in common. Which brings me to the next point.
2. Scaling without compromising security
Every single deployment that moved from pilot to production had one thing in place: a secure, controlled AI infrastructure that the organization trusted.
Not just an off-the-shelf API call to OpenAI or Anthropic.
What they did instead:
A small team of data scientists, engineers, and domain experts worked directly with the AI provider (Mistral AI in this case) to build and fine-tune a custom model:
- Trained on their own proprietary data.
- Embedded with their tacit knowledge and business context.
- Deployed in an environment they could control.
This is especially true in Europe, where regulations are tight. But I think it applies everywhere.
It's not enough to plug any foundation model into your system and hope for the best.
The use cases that actually work treat proprietary data as a competitive advantage. They invest in the infrastructure first.
3. Small, specialized models are quietly becoming the future
Alongside this, there's a broader trend worth paying attention to: domain models.
These are small language models (SLMs) fine-tuned on specialized knowledge. Think:
- a finance model trained on years of actual investment decisions and reasoning chains, built to give precise, metric-aware analysis. Not broad, shallow summaries.
- a code translation model that converts legacy COBOL into Python to help modernize the organization's codebase.
When trained well, these smaller models are more efficient, more consistent, and more accurate than a general-purpose model for their specific task.
What's more: the organization that trained it owns it. No dependency on an external API. No vendor lock-in.
This is still early. But it's the direction things are heading.
4. Not everything causing noise deserves your attention
At the lunch break, I talked to a start-up founder. She co-founded a company that provides AI-powered translation APIs.
I asked how she keeps up with the pace of everything in the AI space (I was half-hoping she'd share some secret system! Btw, have you heard Claude's Opus 4.8 just dropped??)
What she said stuck with me: "Just like everyone, I can barely keep up with the latest AI news. But most of it is distraction. If I don't see a clear use case or benefits of these new things for what we're building, I just don't care. I haven't even tried OpenClaw yet! We're a team of 8. If one of us gets distracted by every shiny new thing, we lose momentum."
I loved how pragmatic she was. No FOMO. No performance of staying current.
And I think this is just as true for individuals as it is for companies.
If you've been feeling guilty for not having tried X or Y...
Well, you DON'T have to!
For me personally, it's far more important to get clarity on your goals and what you're actually working toward. And focus on what's important to get you there.
That clarity is genuinely harder than it sounds. And it matters a lot more than keeping up.
What's your take on these? I'd love to hear which one resonated most.
Hit reply and let me know!
Chào,
Thu
