Keynote at Proxify Morning Bytes: Practical Learnings from Productizing LLM Systems
I had the honor of delivering the keynote at Proxify’s Morning Bytes event in Stockholm, sharing practical learnings from productizing LLM systems at CSPaper.org. [Slides PDF]
The session attracted a full house of tech leaders eager to discuss the realities of building AI systems in production. Between guests stuck in the elevator and fire trucks pulling up (yes, really!), it turned out to be a morning to remember.
Key Takeaways
Two insights that resonated with the audience:
1. More is not always better. Piling on more parameters, more data, or more powerful models doesn’t guarantee success — and can even backfire. By truly understanding the problem and choosing the right data, tools, and intelligence, we can build agents that are both more efficient and deliver better results.
2. Humans at the center. AI creates the most value when combined with deep human expertise. The better we understand the challenge, the more we can amplify the effect of AI models.
Practical Advice for Creating with AI
- Design so use feels natural — no manual needed
- Partner with UX/UI designers
- Bring in domain experts
- Support safe exploration — let users try, fail, and learn
- Deliver fast, deliver small, then iterate with real users
A huge thanks to Viktor Jarnheimer, Danieli Rohr, Paula Avellana, Philip Henriksson, and the entire Proxify team for the warm hospitality and creating a space where connection felt as important as the content. Their attention to detail — from taking coats to refilling coffee cups to adjusting the screen mid-session — made the crowd feel right at home.
As one attendee put it: “Truth is like poetry,” overheard in a bar, somewhere in D.C.
Looking forward to the next one!