Thinking in Platforms

Kaspar von Grünberg and Luca Galante — co-authors of the new book "Thinking in Platforms" and co-founders of Humanitec and the PlatformEngineering.org community — trace their field back to an origin story most engineers will recognize: a company held hostage by a single backend engineer who could threaten to walk unless he got a 3x raise. That "key person dependency" problem, they argue, is exactly what industrialization solved a century earlier when Henry Ford moved manufacturing from artisan workshops to standardized assembly lines — and it's the same move platform engineering makes for software: treating the internal setup as a product instead of something that "just happens" to a team. The book is structured in three parts — historical foundations, a "path-to-outcome" conceptual framework built around golden paths, and a hands-on playbook for going from a minimum viable platform to full enterprise adoption — because, as they note, most platform efforts don't fail on the technical side; they fail on the social side, in getting buy-in from developers, executives, and security teams alike.

The conversation's sharpest point for engineers building with AI right now: agents make platforms non-optional, not less necessary. An AI agent dropped into an enterprise full of undocumented "tribal knowledge" fails for the same reason a new hire would — except at scale, and fast. Von Grünberg's analogy: it's like hiring a brilliant intern and sending them into an office where nobody speaks their language, then being surprised there's no output a week later. That means real infrastructure — context management, security, observability, governance — has to exist before agents can be productive, and the old rule of thumb that you needed 100+ developers to justify a platform team no longer applies when a single person can be running a thousand agents. The bigger claim: platform engineering is already spreading past IT into security, data, and eventually every knowledge-work function, becoming less a software discipline and more the operating model for AI-first organizations.