What they did
Amit Shrivastava wanted to compare the twenty-plus AI plans that cost around $20 a month and up. He set up five Claude Code subagents in a public repo: a price scout that finds each plan's current price on the vendor's page, a benchmark analyst, a review reader that collects hands-on reviews, a value ranker that scores plans and writes a data file, and a fact checker that reopens every source and can only pass or fail the work. The three research agents run on Sonnet to save usage. A slash command takes a budget and a kind of work, such as coding or writing, and another refreshes every price. Rules: no prices from memory, every price needs a link and a date, and no padding the list.
How it went
The first run on 16 November 2026 covered 27 plans at US list prices and produced ranked lists for coding, writing, design and research at budgets from $20 to $100, shown on a static web page. At $20 the top three coding plans landed within 0.2 points of each other. Six tools, including Grok and Cursor, were marked uncertain because sources disagreed after recent plan changes, and some budgets returned only six or seven confirmed plans instead of ten.
Worth knowing
Scores are the agents' judgement from linked benchmarks and reviews, not measurements, weighted 60% quality and 40% usage limits. It runs on your own Claude plan's allowance. Prices move fast, so check the date and the vendor's page before you pay.