In April I started playing round with generating a semi-automated pipeline with python and Claude to generate a private podcast that does a deep dive into AI research papers. I think it is really cool. It fetches papers, scores them based on novelty, importance, relevancy, etc., and then writes the podcast script. It then generates the show by using Eleven Labs voices and then puts it in an RSS feed that Apple Podcasts is ok with.

My personal expense for generating this stuff is a sunk cost, so in May I opened it to everyone via https://paperdive.ai/ It was really fun working with Claude 4.7, 4.8 and even Fable 5, to make the site and refine the pipeline. Now I put out about 4-5 episodes a day. I listen to each one first, then promote the "staged" episodes to my "prod" feed.

This weekend I just add weekly and monthly reviews. Claude will write the review and another instance will generate a script the converts the review into something (hopefully) easier to listen to.

Here is last week: https://paperdive.ai/review/weekly-2026-06-14.html And here is the review for May: https://paperdive.ai/review/monthly-2026-05.html

Now, during my commute, I listen to the individual episodes and the reviews - I typically run them at 1.5x to 2x speed. I have become a much better user of the AI tools by keeping up with recent research.