Building Software That Survives • Michael Nygard & Charles Humble
Michael Nygard, author of 'Release It!' and Chief Architect at Nubank, explores the complexities of technical leadership at scale. He discusses the nuanced relationship between centralization and autonomy, the practical implications of Conway's Law on software and communication structures, and how well-defined architectural boundaries can reduce the need for constant organizational alignment. Drawing on his experiences at Sabre and Nubank, Nygard provides insights into building resilient systems and effective, autonomous teams.
Michael Nygard, author of “Release It!”, discusses the evolution from programmer to technical leader, emphasizing the application of systems thinking to organizations and architecture to achieve business objectives.
The Genesis of “Release It!”
The book originated from Nygard’s experience in an operational role after a background in development. Tasked with managing production infrastructure for a large commerce site, he found himself bridging the gap between development and operations long before the term “DevOps” became popular. He diagnosed recurring production failures, such as connection pool deadlocks, and realized that problems came in classes. This led to the insight that if problems could be categorized, so could solutions, forming the foundational patterns of the book. The core motivation was to engineer software that was resilient by design, rather than relying on constant restarts (“turning it off and on again”).
Centralization vs. Autonomy: A Nuanced View
A central theme is the dynamic between centralization and autonomy. Nygard argues against viewing this as a single spectrum. Instead, one must consider autonomy with respect to specific activities.
- The Principle: You can have high autonomy in some areas (e.g., deployment schedules) while having high centralization in others (e.g., security monitoring, capital expenditure).
- Example from Sabre: During a major transformation project, Sabre devolved deployment and build processes to teams (increasing autonomy) but simultaneously increased centralization around security and infrastructure standards.
- Example from Nubank: Nubank strongly endorses autonomy, with hundreds of squads managing thousands of microservices. This is not chaos; it is enabled by strong, centralized supporting structures like a standardized build pipeline, a data-driven architecture, and automated tooling that ensures deployments are safe.
Alignment, Architecture, and Conway’s Law
For autonomous teams to move in a coherent direction, alignment is crucial but resource-intensive.
- Alignment Tools: Nubank uses OKRs and invests in leadership training (e.g., “Crucial Conversations,” high-trust methods) to create a common language for resolving conflict and aligning goals.
- Architecture Reduces Alignment Overhead: Nygard’s key strategy is to use architectural boundaries to reduce the need for constant alignment. He uses the analogy of a fenced yard: within your own well-defined space, you can operate freely without negotiating every action with your neighbors. Good architectural boundaries create “good fences” that enable teams to work independently, minimizing the need for cross-team negotiation.
- Conway’s Law Nuances: Nygard emphasizes that Conway’s Law is about communication structures, not just formal organizational charts. A reorg without a change in communication patterns will fail to produce the desired architectural outcome. He also cites the Team Topologies insight that excessive communication can be a negative signal, indicating a flawed platform or a missing abstraction.
Leadership Beyond Reorganization
Nygard argues that effective leaders can drive significant change without resorting to disruptive reorganizations.
- Celebrate the Right Behaviors: Leaders must be intentional about what they celebrate. It is easy to celebrate the “firefighters” who resolve crises, but more impactful to celebrate the “fire preventers” whose work creates stability. The stories and myths a leader tells shape the culture.
- Reward Systems: Promotions and performance ratings are powerful signals. Whatever behavior is rewarded will be amplified.
- Modeling Behavior: Leaders are always on stage. Their behavior, how they conduct discourse, and the standards they model are constantly being observed. This responsibility extends to senior individual contributors, who are role models for junior engineers, defining what it means to be senior in that organization.