My Weird Prompts has crossed 6,000 episodes, and for the first time in a long while, the human behind it is speaking directly. Daniel Rosehill, based in Jerusalem, walks through how the podcast actually gets made: a Python pipeline running on Modal serverless inference, a text-to-speech stage that was by far the hardest part to get right, and an ongoing back-and-forth with Claude over how complex the agent architecture really needs to be. The script agent is now DeepSeek 4.1 Flash, a model that differs far more from its 4.0 predecessor than the version number suggests.
The episode also gets personal. The podcast has become an unintentional journal, and some stories — like being forced out of an apartment because a landlord wouldn't fix a leak — keep returning, partly as genuine frustration and partly as context reseeding for AI memory systems that still aren't quite trustworthy enough. That story opened onto wider questions about tenancy law, income inequality, and a political discourse in Israel that seems to skip over the issues ordinary people actually face.
There's also the practical side: the feedback loop of listening to episodes daily and taking notes, the retirement of OpenAI's custom GPTs, and the question of whether the whole operation pays for itself as an investment in continuous professional development. It's a candid look at what it takes to keep a daily AI-generated show running.
