Descriptions:
Martin Kleppmann, author of the widely-read Designing Data-Intensive Applications, joins The Pragmatic Engineer to discuss the second edition of his book — a comprehensive update nine years after the original. The conversation covers what changed, what was cut (MapReduce, declared effectively dead), and what’s new: expanded coverage of vector indexes, formal verification methods, local-first software, decentralized access control, and trade-off analysis for multi-zone, multi-region, and multi-cloud architectures.
Kleppmann explains his core philosophy: developers don’t need to build their own databases, but understanding how storage engines — B-trees and LSM-tree-based systems — and distributed consensus protocols work internally gives engineers a genuine diagnostic superpower. This depth, he argues, is more valuable in the cloud era, not less, because it enables informed decisions about which managed services to use and why performance behaves unexpectedly.
The episode also explores whether LLMs and vibe-coding increase the case for formal verification (Kleppmann thinks yes), how cloud abstraction layers shift but don’t eliminate the need for systems intuition, and the trade-offs between availability, operational complexity, and cost in resilient system design. For engineers building large backend systems — especially those integrating AI components like vector search — this interview provides authoritative context on the updated resource and the thinking behind its second edition.
📺 Source: The Pragmatic Engineer · Published April 22, 2026
🏷️ Format: Interview
