Notes on system design, engineering decisions, and lessons learned building real products. Some are polished evergreen content, others are ideas still taking shape.
For the last two decades, many software careers began with CRUD (Create, Read, Update, Delete) applications. A junior developer would receive a ticket to build a user form, add database persistence, create an API endpoint, add validation, and render a table view. They would then repeat this cycle
What actually happened is that the AI showed up on Monday asking where the repository lives, spent Tuesday reading the entire codebase, and by Wednesday was submitting pull requests while openly questioning architectural decisions made by senior engineers. The latest models don't just "write a function."
I spent a week running Gemma locally, comparing it against Llama, Qwen and GPT-4o, trying to answer one question: Is Google's open model strategy finally good enough for real production workloads?
Software rarely dies from bad ideas. Bad ideas are easy to spot; they get laughed out of the room. Software dies from too many *good* ones
A founder’s app had glowing reviews and enthusiastic users, yet nobody was actually using it. The brutal truth only surfaced when he stopped asking for praise and started asking for a credit card, revealing the hidden danger of free validation
Every failed project is full of sincere people who promised to “do better next time.”