Exactly a year ago a fire broke out in Narbonne, the twin town of my chosen home, Weilheim. The heatwave in France sparked an idea: a simple app that shows cool spots nearby. That idea evolved into Talk2Earth, a talk‑to‑data solution that makes Earth observation (EO) data available to anyone, anywhere, at any time for any use.
After convincing ESA, we had to move quickly, because just two months later they wanted to showcase Talk2Earth.com at the ESA Council Meeting in Bremen to ministers from the member states and the EU. Our team consisted of two developers and a project manager, with no established best practice for a chat‑with‑your‑data solution.
We allocated four weeks to prototype and another four weeks to implement, delivering the app in time for the Council Meeting. During the prototyping phase we tried and discarded many ideas, learning a great deal about the design of multi‑agent systems.
A few weeks after Talk2Earth launched, Rohit Agarwal, a frequent speaker at Machine Learning Week Europe, asked whether I would read a pre‑print of his new book “Beyond The AI Demo: Enough of POCs. Let’s Talk Production“. When I began reading, I was surprised to find Rohit using a talk‑to‑data solution as example—precisely the challenge we faced with Talk2Earth!
Had we read the book beforehand, we probably would have saved a month of prototyping. That is why I recommend the book, with my foreword:
“Developing agentic AI solutions is the easy part. The hard part is operating these systems. Security, safety, reliability, responsibility, scalability and efficiency are all requirements that often contradict each other.
As an AI engineer, it is your job to find the right balance. In his book Beyond the AI Demo, Rohit Agarwal provides the right framework for designing multi-agent systems that excel in operations. Starting with a simple chat-with-your-data application, he demonstrates the trade-offs within his 4 R framework for agentic design: Reliability, Respect, Reach and Restraint.
To strengthen one or more of these four dimensions, he introduces design patterns, gradually increasing the complexity of the solution chapter by chapter. This enables you to learn about different architectures through a real-world example, as well as how to assess their respective pros and cons.
It is a real pleasure to follow Rohit’s journey from start to finish, and his in-depth exploration of technical details that make a difference. Having known Rohit for many years as an excellent and frequent speaker at Machine Learning Week Europe, I can say that this book reflects the pragmatic and progressive mindset of an AI expert and explorer.
If you want to exploit the potential of Agentic AI, this book is your travel guide to this new territory.”
Well done, Rohit! I hope your next book comes out sooner. 😉
