Index
Don’t be a meat proxy – It’s becoming increasingly easy to just pass AI output from one person to another without really understanding it. This post is a good reminder that reading, validating, and rewriting AI-generated answers in your own words is still an important part of the work, and also a matter of respect for the people who have to read, review, and act on what you produce. Otherwise, you’re just becoming a “meat proxy” for the model.
Why some people mow a lawn better than others – Mowing a lawn effectively is a surprisingly interesting optimization problem, how do you find a path that covers the entire area efficiently? It’s a fun example of how everyday problems can hide interesting algorithmic challenges, this post turns out into a amazing data structures and algorithms learning.
LLMs reward expertise – LLMs are making it easier for everyone to become a generalist, but domain expertise may actually become more valuable, not less. This post explores how people with deep knowledge can get significantly more out of the same model.
Software for One – With the decreasing cost of code, the option to code software for small group of people (or even one) become more reliable, if you need a specific software just for you, now is cheaper to build and maintain than ever. I think in the future we’ll see even more examples of it.
MISC
The new rules of context engineering for Claude 5 generation models – With new generation models, some old rules that we used to follow for context engineering are changing, sometimes, less prompting can produce better results.
The Productivity Mirage – Sometimes we can spend so much time trying to find ways of improve our productivity than actually focusing on solving the right problem. This short story about a prolific Facebook engineer remind us that.
How to Make a Nintendo 64 Game in 2026 – Building a game for a 30-year-old console is a crazy engineering challenge. The post goes deep into the challenges the author faced to make it possible to release a game for this console nowadays.
Nothing Works and Everyone Is Euphoric – We have better tools than ever to build software, even with heavy usage of AI models to generate code, software seems to be getting worse. This post explores why AI-driven productivity hasn’t translated into better products yet.
O. Valadares' Blog 