Embracing the 3-Month Rule: A Practical Approach to Non-Scalable Coding

In the tech world, Paul Graham’s mantra of “doing things that don’t scale” is well-known, but its application in coding often goes undiscussed. After dedicating eight months to developing my AI podcast platform, I’ve created a straightforward framework centered on a concept I’m calling the “3-Month Rule.” Essentially, I give each unscalable hack a trial period of three months. If it proves its worth, I invest in a robust solution; if not, it gets phased out.

As engineers, we are often conditioned to craft scalable solutions right from the starting block. With a focus on design patterns, microservices, and distributed systems, we aim to develop architectures that can accommodate millions of users. However, this perspective can lead to premature optimization at startups, where focusing on potential future needs may simply stall progress. My 3-Month Rule encourages the development of straightforward, albeit “imperfect,” code that can be deployed quickly, allowing me to truly understand user needs.

Insights from My Current Infrastructure Strategies

1. Centralized on a Single Virtual Machine

I operate everything—database, web server, background jobs, and even Redis—from a single $40-a-month VM. There’s no redundancy and backups are done manually to my local machine.

What seems reckless here has actually been quite illuminating. Within just two months, I gained valuable insights into my actual resource requirements—knowledge that far surpasses what a standard capacity-planning document could offer. For instance, my “AI-heavy” application only peaks at 4GB of RAM. The complex Kubernetes setup I once considered would have simply meant managing idle containers.

When the system experiences downtime (which has happened twice), I gain useful data on failure points, revealing unexpected vulnerabilities.

2. Configuration Hardcoding

Here’s the way I handle configurations:

python
PRICE_TIER_1 = 9.99
PRICE_TIER_2 = 19.99
MAX_USERS = 100
AI_MODEL = "gpt-4"

Instead of using config files or environment variables, I rely on constants that are scattered throughout my codebase. Adjusting a value necessitates redeployment, but this approach has a hidden advantage: I can search for any configuration with a simple grep command and track every price change in my git history. The time invested in forming a dedicated configuration service would have