Blog

After using Claude Code on projects for a while, the biggest friction isn't the AI getting things wrong - it's the AI doing too much. Think of your CLAUDE.md/AGENT.md as a constitution for how AI behaves in your codebase, and Karpathy's rules are a solid starting point, they keep the agent focused, surgical, and honest, which means cleaner diffs and fewer "why did it touch that?" moments in review.

If you're serious about integrating AI into your business - not just talking about it - this Stanford CS230 lecture from Kian Katanforoosh is required watching.

It cuts through the noise and gets practical fast: how to write prompts that actually work, why RAG beats fine-tuning almost every time, and how agentic workflows are changing what businesses can automate. Real examples, real tradeoffs, no hand-waving.

Before you reach for a fancier model, get more out of the one you have. Better context. Smarter prompts. The right architecture for the job. This is the kind of foundational thinking that saves you from making expensive decisions based on hype.

Watch it. Take notes. Thank me later.

The awesome team at Tighten published a great document on how to implement RAG in a Laravel project with semantic search and vector databases.