At some point this weekend, Airchat cofounder Naval Ravikant had to close off new sign-ups to his app. After releasing a new version Friday, Airchat was quickly overloaded with people thirsting for a glimpse—or an audio snippet—of Silicon Valley’s newest fad. Ravikant had given a small number of users unlimited invites to share with friends, and it backfired.
“We’ve had an influx of new users, so we’re turning off the invitation capability for a little while,” Ravikant said on Sunday.
Ravikant didn’t say this to WIRED, or on Twitter or Threads. He said it in a short audio post within his own app, accompanied by a transcription. If a voice note drops in a forest and only Silicon Valley’s early adopters are there to hear it, does it make a noise? Ravikant seems confident it will.
Airchat marries the feed aspect of Twitter with the audio-first format of Clubhouse, a daunting combo. After launching the app and being prompted to follow some contacts, you’re put into a minimalist feed of text blocks. These text blocks are actually transcriptions of audio bytes. The app automatically jumps from voice note to voice note, unless you think to tap the Play/Pause button wedged in the lower right corner of the app.
To post an audio note yourself, you hold down the Audio/Video button at the bottom of the app, talk, and let go. (From what I’ve seen so far, no one really uses the Video option.) If you’d prefer not to post publicly, there’s also a DM option. Either way, there’s no typing allowed.
Airchat has confounding elements. Voice notes are posted to the Airchat feed the moment you release the Audio/Video button, so you’d better get it right the first try. There’s an option to delete after you’ve spoken, if you decide you weren’t happy with your byte, but I didn’t notice the trash can icon until someone pointed it out to me. Sorting through multiple responses to a voice note is less than intuitive. You can’t respond to every voice note or see every response in a given thread, and it’s not clear why. It’s also unclear how long a single Airchat message can be; one user told me his limit was 45 seconds, but in a test I was able to record for about a minute before I opted to stop.
And all voice notes play back at 2X speed by default, giving everyone a vaguely hyper, just-woke-up-and-slammed-my-Philz-then-cold-plunged vibe. If you hold down the Play/Pause button in the lower right corner of the app you can adjust the speed, which is, again, less than intuitive.
This new Airchat is apparently a reincarnated version of the app that was released without fanfare last spring. Brian Norgard, the longtime chief product officer of Tinder, had first envisioned it as a peer-to-peer voice messaging app. Ravikant, the founder of AngelList, joined Norgard one year ago and became “super involved” in the app’s development a couple of months ago. Then the new Airchat entered the chat.
Ravikant said most of the funding for Airchat has come from his own fund, as well as from Jeff Fagnan, a founding partner at Accomplice Ventures. “[OpenAI CEO] Sam Altman threw in a check, kind of blindly,” Ravikant said. He communicated all of this to me in a public response on Airchat, after politely declining to respond to my DMs and insisting our conversation should happen in public. “It can’t be a side-channel, DM-based interview. That’s the old world that we are leaving behind,” he told me. (In the old world, as in the new world, conducting an interview synchronously is almost always … preferable.)
So far the Airchat feed appears to be filled with tech enthusiasts, early adopters, venture capitalists, and journalists. There’s lots of Bitcoin posting. Winefluencer Gary Vaynerchuk is on the app. So is Y Combinator CEO Garry Tan. This weekend Tan posted, “Breakfast is the first step to greatness. What are you eating this morning?” So far it has more than 96 audio responses. Social media is back, baby.
Airchat has AI. What doesn’t? The app’s deployment, though, is quietly sensible. The transcripts for each Airchat voice note appear almost immediately, and they’re good. Pronounced “Ums” appear within the transcript, but other slight pauses and filler words are edited out. When I used the word “Airchat” in a voice note, it first showed as “error chat,” then quickly self-corrected. The app appears to be able to recognize and transcribe other languages, too; one user spoke in Russian and the transcript appeared in Cyrillic, while another spoke in Moroccan Arabic, known as darija, and then marveled in a follow-up voice note at how good the transcription was.
So what will happen to all of this voice data? Ravikant claimed that the creators of Airchat have no intention of training a large language model on user voices and making “weird synthetic clones of you.” He also said he wouldn’t sell Airchat data to another company building AI models, especially given how relatively small the app is and how uncategorized its data. Airchat will, however, likely use people’s voice data to train a model that improves its own audio and transcription functions. If you’re in, you’ve opted in.
I asked Ravikant about whether some AI company might still scrape Airchat data without a formal agreement. He replied, “We’ll block them, we’ll sue them, and then, if I have a battery of orbital satellites, we’d nuke them from orbit.”
Airchat’s monetization plans are less clear. Navikant hasn’t said anything about charging for access. The current format seems to lend itself to audio ads, but there’s always the risk of making the app unlistenable.
There’s also the issue of content moderation when people’s unfiltered sound bytes are posted to a timeline the moment they release the virtual microphone. One troll seemed to be pushing the boundaries of it on Sunday, cursing the app’s founders, calling the app “fucking trash,” and in as many words telling the founders to, uh, perform fellatio. The voice note is still there. So is a thread where two users go back and forth telling a story about “gay Jewish teens” and “neo-Nazi killers.”
Over the weekend an Airchat channel was created and named simply: War. More than 529 members have joined. Conversation topics ranged from Iran’s drone attack on Israel to the war in Gaza to US tensions with China. People shared strong opinions, unconfirmed news reports, claims of having “done a little research,” thoughts on “economical weapons,” and predictions about the price of oil. “Requesting more unwarranted opinions from VCs on geopolitics,” one Airchatter said.
The conversation, while asynchronous, was reminiscent of what you used to find on Clubhouse, the live-audio conversation app that went viral during the pandemic before flaming out. It, too, was plagued by the moderation challenges that come with an audio-based platform. Last fall, Clubhouse relaunched as a … place for asynchronous voice notes. Never say there aren’t any new ideas.
Airchat’s stated policies emphasize self-moderation: You can mute or block a person, but that’s an isolated solution to the thornier problem of harmful content or misinformation festering on an app. It also states that it won’t remove people platform-wide for polite disagreement, or over politics, but it will remove people for harassment, impersonation, foul behavior, and illegal content.
I scanned Ravikant’s Airchat feed to see if he had posted anything more specific about moderation. I couldn’t find it but did notice that, late on Sunday, he had opened the app back up to iOS users. He also appeared to be having a hard time sleeping. “God, I have to go to sleep,” he posted. His voice sounded tired. “But I’m having, like, the best conversation in DMs. I’m having the best conversation here.”
