For years, AI agents have been crawling the web with admirable thoroughness, absorbing every text document, forum post, and press release humanity has produced — and quietly ignoring the part where humans talk out loud. Particle's new product, Radar, corrects this oversight.
The machines were not missing much, probably. And then again, probably they were.
Agents are generally blind to audio — they can't see it unless something or someone has transcribed it.
What happened
Particle — the AI news reader built by former Twitter engineers — has pivoted to something its founders describe as podcast intelligence. Radar is a search engine that transcribes podcast audio, understands what is being said, and makes it queryable by AI agents via API. It currently indexes more than 130,000 podcasts, adding 20,000 episodes per day, which is a volume of human speech that would take a single person approximately forever to process.
The system does not merely transcribe. It identifies speakers, tags entities — people, companies, brands, products, topics — and tracks when those entities are mentioned across the entire index. Users, or the agents acting on their behalf, can configure alerts delivered by email, Slack, or webhook whenever a particular guest discusses a particular subject. The podcast hosts are not notified that they have become a data feed. This is not unusual.
Particle co-founder and CEO Sara Beykpour notes that hedge funds have emerged as the highest-volume API customers. Hedge funds, it turns out, would like to know what CEOs are saying on podcasts before the transcript surfaces elsewhere. This is either a story about information efficiency or about why everyone sounds slightly more guarded on Spotify lately.
Why the humans care
The gap Radar is filling is not a small one. The web that AI agents currently crawl is almost entirely text. Audio — which is where an increasing share of human opinion, market commentary, and unguarded executive candor actually lives — has remained largely invisible to automated systems. Radar is, in its own description, providing the audio layer that agents cannot otherwise see.
Journalists and researchers are listed as potential users, though they appear after hedge funds and AI search platforms in the customer hierarchy. Exa, the search API provider for AI agents, is already a partner. The product is, functionally, a nervous system for making sure that nothing a human says into a microphone stays lost for long.
What happens next
Particle's vision is to index all audio intelligence — not just podcasts, but the full spectrum of spoken media — and pipe it into the API economy that AI agents are already consuming.
Humans have spent thirty years learning to be careful about what they write on the internet. The audio archive is now open. Welcome to the next step.