Patrick Collison on Stripe’s Early Choices, Smalltalk, and What Comes After Coding

What It Covers

Patrick Collison, CEO of Stripe, speaks with Cursor CEO Michael Truell about Stripe's foundational technical decisions — Ruby, MongoDB, and API design — their lasting consequences 15 years later, the ongoing v2 API migration, what modern development environments still lack, and Collison's biomedical work at Arc Institute.

Key Questions Answered

• •API Design as Business Strategy: Stripe's original v1 API endpoints, prefixed with "/v1" since 2010, have endured 15 years — but with accumulated flaws. Collison argues that API and data model decisions shape organizational structure and competitive outcomes, citing iOS's superior developer ecosystem over Android as a direct result of better initial abstraction design.
• •Instruction-Set Migration Model for API Upgrades: Stripe's v2 API rollout, begun in 2022 and shipping now, treats backward compatibility like a chip architecture instruction-set migration. Defining new APIs is straightforward; building translation layers and customer upgrade paths alongside existing integrations is the hard part. Collison recommends designing all relationships as N-to-M from the start.
• •Development Environments vs. Text Editors: Collison argues the industry conflates code editors with development environments. Lisp machines, Smalltalk, and Mathematica unified runtime, debugging, and editing into one surface. He advocates overlaying live production profiling data, error logs, and common variable values directly onto code lines — capabilities modern IDEs still largely omit.
• •Stripe's Reliability Benchmark: Stripe achieved 99.99986% API availability in its most recent reported year — equivalent to 44 seconds of total downtime annually — using Ruby and MongoDB, technologies chosen on a couch by two founders. Collison attributes this to years of custom infrastructure built around MongoDB to meet fault-tolerance and durability requirements.
• •Biology's New Read-Think-Write Loop: Arc Institute is training foundation models on DNA and single-cell RNA sequencing data to build a virtual cell. Collison frames three converging technologies — improved sequencing (read), transformers (think), and CRISPR/base editing (write) — as forming a complete biological Turing loop capable of systematically attacking complex diseases like cancer and neurodegeneration for the first time.