Embracing the 3-Month Rule: A Pragmatic Guide to Scaling Your Startup
In the startup world, the well-known mantra of Paul Graham, “Do things that don’t scale,” often creates confusion—especially among developers. How do you truly apply this advice within the realm of software development? Drawing upon my eight-month journey in building an AI podcast platform, I’ve formulated a straightforward approach: every unscalable tactic is permitted to exist for three months. After that, it either justifies its existence with proven value or is phased out.
As engineers, we are frequently conditioned to focus on creating scalable solutions from the outset. The allure of elegant design patterns, microservices, and robust distributed systems beckons us, envisioning a future bustling with millions of users. However, in the realm of startups, striving for scalability can often lead to costly delays as we inefficiently prepare for customers that don’t yet exist. My three-month rule compels me to produce straightforward and occasionally “imperfect” code that progresses to production, providing invaluable insights about user needs.
Current Infrastructure Hacks: Challenges Turned Insights
1. Simplifying with a Single Virtual Machine
My entire stack—database, web server, background tasks, and Redis—operates on a single, low-cost VM. While it lacks redundancy and relies on manual backups, here lies the brilliance: in merely two months, I’ve gained clearer visibility into my actual resource requirements than any intricate capacity planning document could offer. My “AI-centric” platform now reveals that it typically requires only 4GB of RAM. The complex Kubernetes architecture I nearly implemented would have resulted in managing idling containers.
Whenever it crashes—twice so far—valuable data is retrieved on the true points of failure, which are often surprising.
2. Hardcoded Configuration as a Learning Tool
My configuration values are hardcoded directly into the codebase, eliminating the use of config files or environment variables. Changing a parameter necessitates a redeployment, and this has turned advantageous: I can swiftly search my entire codebase for configurations using grep. Each price modification is stored in git history and reviewed (albeit by me during my own pull requests).
Setting up a configuration service could take a week, yet I’ve only modified these constants three times over the last 90 days. This translates to 15 minutes of redeployment as opposed to 40 hours of work.
