In this week’s Goodread, author Nicole Koenigstein delivers a strong closing statement: “The winners in the agentic era won’t be those with the smartest instructions but the ones who build the most resilient collaboration structures. Agentic performance is an architectural outcome, not a prompting problem.”
With the development of Talk2Earth.com—an agentic AI solution for EO data—as part of ESA’s “European Resilience from Space” (ERS) program, we can back up this statement. It’s not about prompt engineering, but system design—or more precisely, multi-agent system design.
Btw: check our free template for Multi-Agent System Design on Miroverse!
Nicole makes a valid point: “Papers on agentic and multi-agent systems (MAS) skyrocketed from 820 in 2024 to over 2,500 in 2025. […] Yet there is a disconnect: While research is booming, these systems still frequently fail when they hit production. [..] I use the term prompting fallacy to describe the belief that model and prompt tweaks alone can fix systemic coordination failures. You can’t prompt your way out of a system-level failure.“
She further argues: “There is no universal best [system design] pattern, only patterns that fit the task and the way information needs to flow:
⚡A linear, supervisor-based architecture is the most common starting point.
⚡ In creative settings, a blackboard-style architecture with shared memory often works better.
⚡ In peer-to-peer collaboration, agents exchange information directly without a central controller.
⚡ Swarms work well in tasks like web research because the goal is coverage, not immediate convergence.
Considering all of this, many production systems benefit from hybrid patterns.”
And she has a simple advice: “I design AI agents the same way I think about building a team. Each agent has a skill profile, strengths, blind spots, and an appropriate role.”
Read the full Goodread of the week:
👉 https://www.oreilly.com/radar/designing-effective-multi-agent-architectures/
