Why did Stripe build Sorbet? (~2017).

strategy (37), eng-strategy-book (25)

This is a work in progress draft.

Hi

Finally, I think it’s intellectually important to recognize the limits of my own involvement in the Sorbet strategy. While I ultimately led the organization that built Sorbet at Stripe, my organization did not include the Developer Productivity team when they decided to build Sorbet. Indeed, I initially disagreed with the approach: I preferred investing in a migration to a strongly-typed language (specifically Golang), rather than spending time on creating and maintaining a typing framework for Ruby.

Ultimately, while both approaches might have worked, the Sorbet strategy did work, and it was probably the better choice because it allowed us to concentrate the investment into a small team rather than spreading the cost across every team within Stripe.

This is an exploratory, draft chapter for a book on engineering strategy that I’m brainstorming in #eng-strategy-book. As such, some of the links go to other draft chapters, both published drafts and very early, unpublished drafts.

Reading this document

To apply this strategy, start at the top with Policy. To understand the thinking behind this strategy, read sections in reserve order, starting with Explore.

More detail on this structure in Making a readable Engineering Strategy document.

Policy & Operation

..

Diagnose

  • developer productivity surveys: Two stages – getting data, sitting with data ← imo this is how you learn; then picking a few specific projects and driving them; together this is the core DX team loop done on a one-off basis
    • imperfect: always something bad, so goal is sort of rotating what’s bad
  • migration of codebase will require many teams learning a new language
  • adding ruby typing will allow a centralized team to make significant investments
  • much of ruby typing can be done via the abstract syntax tree (AST)
  • we can incrementally prioritize typing on highest value places
  • ruby is not a high performance programing language, but it’s generally good enough and we can get sufficient scale at a reasonable price