The week in AI, without the homework

Plain Leverage is a weekly digest about AI that changes what you would build or buy. It is written for operators, small business owners, and technically curious people who do not have a research team and do not want one.

Who it is for

If you run a company, work for yourself, or just find this stuff fascinating, the useful question is never "what did the model score on a benchmark." It is "does this save me time, money, or energy, and is it interesting enough to tell someone about." Every item gets that test before it goes in.

What gets cut

Funding announcements with no product, benchmark charts with no method, server patching, and anything where the only source is the company's own marketing page. If a story cannot be linked to a primary source, it does not ship.

The closing slot

Every issue ends with one item chosen for fascination rather than utility. An emulator baked into a model's weights. An agent that injected a drug into a mouse. If it makes you say "wait, really," it belongs at the end.

How it is made

The engine pulls about 25 RSS and Atom feeds, normalizes dates, removes duplicates, tags each item, and ranks by source weight, recency, and signal. The digest text is drafted with AI from that ranked list. The engine handles collection and ranking. The drafting sits on top of it. The whole pipeline is about 1,400 lines of Python across five modules, with no third-party dependencies.

This publication is also a portfolio piece. I am building it to get better at this work in public, so the method is visible on purpose.

Corrections

If an item is wrong or a link is dead, send a note. Corrections go at the top of the following issue, with the original item named.