Hi there,

Welcome back to Beyond the Prompt! I'm very grateful for the kind messages I got from the first issue. It is making me more motivated than ever to curate the best links for you... so let's get into it.

As the adage goes, "there are decades when nothing happens, and there are weeks when decades happen." I think March 2025 will be one of those pivotal months that will end up shaping history, and yes, our digital lives too.

💡 From the case studies

Just one case study this month, about a topic many organizations handle very poorly: customer segmentation. No matter your sector, B2C or B2B, if you don't have strong customer segmentation, you are leaving money on the table. Simple as that.

That in 2025, so many marketing teams stick to "intuition-based" personas at the expense of data-driven segmentation is... baffling to me. Trying to stay polite here. 🤬

📜 From the blog

If you missed our Operational Efficiency Webinar and have been dying to catch up, fear not. We got the replay and the slides right here.

• We summarized why Fabric is on everyone's mind right now (TL;DR: because it's good enough for most uses).
• We had Loïc tell his story of joining Agilytic. It's the season when young graduates think about their future, and we want to be transparent about what makes Agilytic a special place. If you know a talented young graduate, send them this article 🙏
• We shared why successful proof-of-concepts sometimes fail at the industrialization phase, and what can be done about it.
• Chatbots are in fashion (understatement of the year). Being the pragmatics we've always been, we published a "do's and don'ts" on deploying LLMs in your organisation.
• Document management is a hot topic because the efficiency gains are obvious (see Mistral below). So it makes sense for us to build our own light OCR assets. They make our work much easier — and cheaper for our clients, which is nice.

🎙️ We got a podcast!

You read that right, we're having some conversations in audiovisual formats too. The first episode is out. Head on here for audio, and here for the YouTube version.

"But when do you do some actual work, Agilytic?", you might ask.

To which I'd reply "Most of the time. We're actually very efficient and smart with all of this content stuff. You should hire us. 😇"

🧪 From the lab: What we're cooking

During the last sprint review, two main topics were at the center of discussions in our Lab:

LLM framework standardization and reusability

We created a centralized repository with standardized prompt engineering frameworks, best practices, and reusable code snippets from our recent projects. This ensures greater consistency and efficiency across all LLM projects. We discussed that a key improvement for this framework is containerization to ensure compatibility with production environments.

Agentic AI

You cannot miss it on your LinkedIn feed: AI agents are becoming a super hot topic. But understanding their role, capabilities, and implementation can be a challenge.

This topic will evolve super fast in the coming months, so we will stay ahead of the curve and dive into the world of AI agents.

🌐 From everywhere else

Funny how a simple white house visit can have huge consequences. I wrote about it. Look, I'm not saying you should ditch Office 365 and Azure at once, but in the course of 24h, the topic of sovereign cloud graduated from the "nice white paper, maybe one day" to "if we don't prepare, I might be liable for catastrophic negligence."

Rumors of Google's demise are greatly exagerated. At least that's according to Datos research, highlighting that Google still receives more than 350 times the search volume of ChatGPT. I wouldn't rest on my laurels if I were Google, though.

Microsoft’s new Dragon Copilot is an AI assistant for healthcare.

Remember IBM Watson? Because I do.

Perplexity is doing the Lord's work by open-sourcing a de-censored version of DeepSeek. This opens the door to cheap, EU-hosted, private reasoning LLMs. We will soon experiment with Librechat internally.

ChatGPT might be a better couple's therapist than the average shrink.

Mistral is entering the document sorting game (last time it was Gemini). Document classification is becoming table stakes faster than I can write those damn newsletters. Agilytic has done a lot of document management projects, so we can definitely vouch for their impact. That we can now do this much faster means every company should at least reconsider their document flows. Obvious efficiency gains to find there.

Google is using—you guessed it—AI to suggest calendar events from the contents of your emails. Seems like a minor thing but it probably will become one of those ubiquitous features in mere months.

OpenAI, the company behind ChatGPT, signed a $11.9 billion deal with CoreWeave, your favorite cloud service provider's favorite cloud service provider. Essentially, OpenAI and Microsoft are frenemies on a competitive collision course. The less their destinies intertwine, the less awkward it will be for everyone.

McDonalds is getting into the AI-ops train. As I've said before, 2025 will be the year of efficiency. If AI can prevent the notoriously dreadful ice cream machines from failing, it'll make millions of kids happier in the process.

Matt Ball argues that AI model pickers are a design failure instead of a useful feature, and I tend to agree with him. Having a drop-down to select the right model adds friction, does not help casual users, and, frankly, is ripe for an interface breakthrough.

The Mayo Clinic is working on eliminating hallucinations from LLMs by employing backwards RAG, allowing them to push the models in non-diagnostic areas of the clinic. If there's one industry where you don't want a model to hallucinate, it's probably healthcare.

Apple is having a horrid time with AI and delayed highly-advertised features. It's not usual for Apple to enter vaporware territory (leave that to Google and Tesla), so it's worth noticing.

Turns out having a generic chatbot or generating cute images is relatively easy. Pushing correct, personalized, and useful contextual AI for billions of users? Not so much.

Some suggest they just switch off Siri and let ChatGPT do its thing (IMHO: never gonna happen).

🧑‍💻 Life @ Agilytic

We’ve got a team-building event coming up, which is always a good thing. I’ve always believed that while values start at the top, culture is built from the ground up. I’ve been fortunate to be part of teams where the culture was strong and the camaraderie real. As our team grows, I really hope we keep that going — it makes the work more meaningful and the experience better for everyone.

👋 Parting thoughts

The world is moving fast—tech, geopolitics, markets. It feels like history is on 2x speed... in the wrong direction!

It’s easy to get anxious, but here’s the Stoic take: don’t stress about what you can’t control. Focus on your response.

Marcus Aurelius wrote: “Receive without pride, let go without attachment.” Keep your head clear, your principles intact, and stay sharp. Chaos is a given. Calm is a choice.

All the best,

Julien

_

PS: In case you're wondering about the Ides of March...

PS2: If you can't get enough links from me, I have another one where I post interesting stuff that didn't make the cut here.

PS3: Feedback is a gift. Tell me what you liked (and liked less) simply by replying to this email.