Stripe's Collison shares uptime secrets

Patrick Collison used a February 20 podcast appearance to describe Stripe’s reliability work as the product of old technical decisions that the company still carries. In a conversation with Cursor CEO Michael Truell, the Stripe co-founder said the company still lives with early choices around MongoDB and Ruby roughly 15 years later. (a16z.com) He also said Stripe logged 99.99986% uptime last year, which he described as 44 seconds of annual downtime. The interview, later republished by Andreessen Horowitz, centered on API design, migrations and how Collison thinks teams should build software that lasts. (tldl.io) ### Why did Collison dwell on MongoDB and Ruby from Stripe’s earliest years? Patrick Collison said Stripe’s early bets on MongoDB and Ruby were the kind of choices that keep shaping a company long after the startup phase ends. The a16z episode description says he discussed “the MongoDB and Ruby decisions Stripe still lives with 15 years later,” framing them as examples of how infrastructure choices harden into long-term constraints and advantages. Collison said those decisions still define the company “15 years and 44 seconds of annual downtime later,” according to the episode summary published by TLDL. (a16z.com) That summary says his broader point was that language, datastore and API decisions can carry multi-decade consequences for architecture and cost. ### What does the 99.99986% uptime figure actually tell readers? Stripe’s reported 99.99986% uptime translates into roughly 44 seconds of downtime over a year, according to the podcast summary. Collison presented that figure as evidence of the engineering work required to make a payments API dependable at scale. (a16z.com) The Clypt summary of the same appearance says Collison described Stripe’s API as unavailable for only 44 seconds in an entire year and said the company had to invest heavily to make MongoDB fault-tolerant and distributed enough to support that level of reliability. ### What did he say about rebuilding Stripe’s API abstractions? (tldl.io) Collison said designing the new API surface was not the hardest part of Stripe’s v2 work. The TLDL summary quotes him as saying, “Defining the new APIs is the easy part,” and says he compared the migration problem to an instruction-set change because the new system had to work alongside everything already built on the old one. That framing matches the official a16z description, which says Collison discussed why he would spend even more time on API design if he could do it over. The summary says durable APIs and data models are among the highest-leverage engineering decisions because migrations are expensive and slow once a platform is widely used. (useclypt.com) ### How does this fit with his view of coding in the AI era? Patrick Collison used the same interview to argue for development environments that make debugging and iteration faster. The TLDL summary says he praised interactive Smalltalk- and Lisp-style runtimes where developers can inspect and edit code during execution, rather than relying on a looser editor-plus-runtime workflow. (tldl.io) The official episode description also says Collison and Truell discussed whether AI is showing up in economic productivity data. While the social briefing for this story said Collison warned against LLM “generic slop,” the verifiable public sources surfaced here support the broader point that he was focused on software quality, debugging and long-lived design choices, not on replacing that work with automation. (a16z.com) ### Why does this interview read like an engineering playbook? The February 20 conversation stayed close to operating details: language choice, datastore trade-offs, debugger design, API migrations and uptime. (tldl.io) The TLDL summary says Collison argued that investing more time in API and data-model design is “extremely high leverage,” because those decisions become harder to reverse as a platform grows. The next public reference point is the full podcast itself, which a16z lists as episode 1042 of The a16z Show and notes originally aired on Cursor’s podcast with Michael Truell. Readers looking for the complete discussion can find it there. (tldl.io) (a16z.com)