Embracing the 3-Month Experiment: A Pragmatic Approach to Development

In the world of startups, the advice from Paul Graham to “do things that don’t scale” often circulates, yet very few delve into the execution of this principle within coding practices. Over the past eight months of developing my AI podcast platform, I’ve formulated a straightforward framework grounded in this advice: every unscalable solution is given a lifespan of three months. At the end of this period, the solution either proves its effectiveness and is refined into a scalable version, or it’s phased out.

As engineers, we’re often conditioned to prioritize scalable solutions right from the outset. The allure of design patterns, microservices, and distributed systems can lead us to craft elaborate architectures aimed at accommodating millions of users. However, in the startup environment, this line of thinking can often result in costly delays while we chase hypothetical user needs. My 3-month rule has encouraged me to embrace a more direct approach, allowing me to deploy straightforward, albeit rough, code that can be released quickly and reveals genuine user requirements.

My Current Infrastructure Hacks: Smart Choices in an Unscalable Framework

1. Consolidated on a Single Virtual Machine
My entire stack, including the database, web server, background jobs, and caching with Redis, operates on a single $40/month virtual machine. This setup lacks redundancy, and I manually back up data locally.

Why is this a clever approach? In just two months, I’ve gained invaluable insights into my actual resource requirements—far beyond what theoretical capacity planning could provide. Recently, I discovered that my “AI-intensive” platform peaks at just 4GB of RAM. The complex Kubernetes architecture I nearly implemented would have merely managed empty resources, while real-time data from system crashes (which have happened twice) has provided invaluable insights—often overturning my expectations.

2. Hardcoded Configuration Values
python
PRICE_TIER_1 = 9.99
PRICE_TIER_2 = 19.99
MAX_USERS = 100
AI_MODEL = "gpt-4"

Instead of relying on configuration files or environment variables, I’ve opted to scatter constants throughout my codebase. This approach means that changing any configurations necessitates a redeployment, but here’s the upside: I can quickly search my entire code for any value, ensuring thorough traceability of any adjustments made. In three months, I