The Daily Stitch.

The open-source AI debate has become a religious war. Strip out the ideology and the question a business actually faces is dull and practical: control, privacy, cost, and who owns the off switch.

Every week a new report declares that AI will create more jobs, or destroy them all. They cannot both be right, and the fight over the headline hides the thing that actually matters.

People argue endlessly about which model is best, and the same model somehow gets called both brilliant and useless. They are usually not judging the model. They are judging the harness around it.

A popular take says coding was never the hard part. As a claim it is half true, and the untrue half is where the money and the difficulty actually live.

People keep asking how good models must get before we stop reviewing their code. The honest answer is never, and the reason is more than caution.

People are handing AI jobs that a plain script does faster, cheaper, and without the chance of deleting your home directory. That is not innovation, it is a downgrade.

The dream of an AI employee that never sleeps is real. What the demos hide is that no single tool is the employee. It is a stack, and the stack is the hard part.

The instinct is to feed an AI agent more context. The people who built the agent keep telling you to feed it less.

Most people ask an AI a question, get an answer, and leave. The same model, pushed differently, would have told them what they actually needed to hear. The difference is direction.

We rebuilt this very site with an AI coding agent. It is fast and genuinely good, and it still cannot decide what "good" means.