Bliki: TwoPizzaTeam

A two-pizza team is a small team that fully supports software for a particular business capability. The term became popular as it used to describe how Amazon organized their software staff.

The name suggests the most obvious aspect of such teams, their size. The name comes from the principle that the team should no larger than can be fed with two pizzas. (Although we are talking about American Pizzas here, which seemed alarmingly huge when I first encountered them over here.) Keeping a team small keeps it cohesive, forming tight working relationships. Typically I hear this means such teams are about 5-8 people, although my experience suggests that the upper limit is somewhere about 15.

Although the name focuses solely on the size, just as important is the team's focus. A two-pizza team should have all the capabilities it needs to delivery valuable software to its users, with minimal hand-offs and dependencies on other teams. They can figure out what their customer needs, and quickly translate that into working software, able to experiment and evolve that software as their customer's needs change.

Two-pizza teams are Outcome Oriented rather than Activity Oriented. They don't organize along lines of skills (databases, testing, operations), instead they take on all the responsibilities required to support their customers. This minimizes inter-team hand-offs in the flow of features to their customers, allowing them to reduce the cycle-team (the time required to turn an idea for a feature into code running in production). This outcome-orientation also means they deploy code into production and monitor its use there, famously responsible for any production outages (often meaning they on the hook for off-hours support) - a principle known as "you build it, you run it".

Focusing on a customer need like this means teams are long-lived, Business Capability Centric teams that support their business capability as long as that capability is active. Unlike project-oriented teams - that disband when the software is "done" - they think of themselves as enabling and enhancing a long-lived product. This aspect often leads to them being referred to as product teams.

The wide scope of skills and responsibilities that a two-pizza team needs to support its product means that although such teams can be the primary approach to team organization, they need support from a well-constructed software platform. For small organizations, this can be a commercial platform, such as a modern cloud offering. Larger organizations will create their own internal platforms to make it easier for their two-pizza teams to collaborate without creating difficult hand-offs. Team Topologies provides a good way to think about the different kinds of teams and interactions required to support two-pizza teams (Team Topologies calls them stream-aligned teams).

For business-capability centric teams to be effective, they will need to make use of each others' capabilities. Teams will thus need to provide their capabilities to their peers, often though thoughtfully designed APIs. This responsibility for such teams to provide services to their peers is often overlooked, if it doesn't happen it will lead to sclerotic information silos.

Organizing people around business capabilities like this has a profound interaction with the way the software for an organization is structured - due to the effect of Conways Law. Software components built by two-pizza teams need well-controlled interactions with their peers, with clear APIs between them. This thinking led to the development of microservices, but that's not the only approach - well-structured components within a monolithic run-time is often a better path.

https://martinfowler.com/bliki/TwoPizzaTeam.html

Created 1y | Jul 25, 2023, 1:50:07 PM


Login to add comment

Other posts in this group

How is GenAI different from other code generators?

How is code generation with GenAI different from more "traditional" code generators? The newest memo in Birgitta Böckeler's explorations of GenAI talks abou

Oct 18, 2023, 8:10:05 PM | Martin Fowler
How is GenAI different from other code generators?

How is code generation with GenAI different from more "traditional" code generators? The newest memo in Birgitta Böckeler's explorations of GenAI talks abou

Sep 19, 2023, 10:10:06 AM | Martin Fowler
TDD with GitHub Copilot

At Thoughtworks, we are strong practitioners of Test Driven Development (TDD). Naturally this leads to the question of how generative AI can

Aug 17, 2023, 8:30:07 PM | Martin Fowler