Accidentally sent my orchestrator agent a voice note on Telegram that I assumed would be transcribed first, as Claude app does. It replied saying it couldn't parse audio files.. but then proceeded to autonomously install an speech-to-text app and then transcribed the original message without any intervention from me, before replying. What a world 🤯
commit 80984a6
i remember Peter Steinberger recounting a similar experience when developing openclaw. it is indeed pretty mindblowing.