Embracing the 3-Month Rule: A Pragmatic Approach to Unscalable Solutions

When it comes to startup culture, one phrase stands out: “Do things that don’t scale,” a piece of advice famously offered by Paul Graham. However, the implementation of this principle in the realm of software development often goes unaddressed. After spending eight months refining my AI podcast platform, I have adopted a straightforward framework: each unscalable tactic is given a lifespan of just three months. If it proves its worth, it receives a more robust iteration; if not, it gets phased out.

As engineers, we are often trained to pursue scalable solutions from the very beginning, focusing on sophisticated architectures like microservices and distributed systems capable of handling millions of users. However, this mindset aligns more with large corporations than with startups. In the early stages, striving for scalability can lead to misguided delays, optimizing for potential users who may never materialize. My 3-month rule encourages me to implement straightforward, even “imperfect,” code that fosters rapid iteration and reveals genuine user needs.

Current Infrastructure Hacks: Smart Decisions in Disguise

1. Unified VM for All Services

My entire infrastructure operates on a single virtual machine (VM) that hosts the database, web server, background jobs, and Redis—all for $40 a month. There’s no redundancy and backups are conducted manually.

What seems like a reckless setup is surprisingly insightful. In just two months, I’ve gained a clearer picture of my resource demands than any capacity planning document could offer. Interestingly, my so-called “AI-heavy” platform reaches its peak at only 4GB of RAM. Had I pursued an elaborate Kubernetes architecture, I would have been busy managing idle containers rather than addressing actual needs. Each crash, which has occurred twice so far, provides valuable lessons about system vulnerabilities—revealing unexpected insights every time.

2. Hardcoded Configurations

Configurations are littered throughout my code as constants:

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

Rather than using configuration files or environment variables, this straightforward approach simplifies the process. While tweaking any setting requires redeployment, I can swiftly search for any value across my codebase and track changes through git history. Over three months, I’ve amended these variables only three times; what would