Notes on tech, philosophy and startups. What I learn from building, and the questions that stay with me.
Results: 3
My job is now crafting tickets that a team of agents can build on their own. Each agent picks up a slice, builds it, and proves it end to end, with nobody stepping in. Here are the rules a ticket has to follow for that to work, and what the reviewers caught that I wouldn't have.
Claude Code · AI · Engineering Process · Developer Tools
A neural network is a function with adjustable parameters. Training measures the error, computes gradients, and nudges the parameters. I walk through it with a toy laptop-price model: loss, gradient descent with real numbers, backpropagation, overfitting, and a 10-line script you can run.
AI · Machine Learning · Neural Networks
Not every AI decision deserves the same amount of computation. System 1 and System 2 are a useful analogy for splitting the work between a fast model, a policy, and a slower investigation. Here is how I think about the split, where it breaks, and why confidence alone should never route a case.
AI · Machine Learning · System Design