Radar indexes 130,000 podcasts for AI agents to examine

Particle launched Radar, a podcast search engine, on 26 August. Radar has indexed more than 130,000 podcasts and adds about 20,000 episodes each day. Hedge funds are its highest-volume customers for the API (a connection that lets software get data). AI agents can find spoken content in all of them, but Descript says the edit is the harder step.

How it works

Radar writes the text of each podcast with speaker labels and data about persons, companies and topics. It can monitor mentions, send alerts and give clips with timestamps. It connects through an API and through MCP (Model Context Protocol, a standard that lets agents use tools).

Descript has an MCP connection too. Through it, an agent can cut text, remove filler words, make highlight reels, add captions and publish a page to share.

The numbers

Ahrefs found that 83.2% of marketers use Claude for content, 75.7% use ChatGPT and 57.9% use Gemini. The source gives no test results for the quality of agent edits.

What is not known

Daisy Zhang, head of growth marketing at Descript, says that story, tone, jokes and sensitive edits must have human judgment. She tells a team to start with one repetitive task where the result is easy to judge.

Insight: Finding is easy to sell, editing is not

Hedge funds are the top API customers of Radar, which sells the find step. Descript sells the edit step, and Zhang says that step is harder. This reads as two markets: find is a data business that persons pay for, and edit is a tool business that must have a person. Leaders can pay for data access first and keep a human sign-off on edits.