@jack Sooo cool to see!! We did this small part of this using Claude code channels feature (we call it fleets) to work along side any sessions and all Claude chats across my computer Harness has the fleet exposed as a tool, this is really nice to see also agent group chat?? So cool!
One pattern I find useful for working with LLMs is a nice long ramble session. Sometimes the LLM needs more bits to understand what you're trying to achieve, but you're too lazy to type them. In these cases I like to lean back, switch to /voice and just ramble for like 10 minutes, total mess, anything goes, full stream of consciousness. Sometimes I declare it up top, something like "switching to speech recognition sorry for any typos...". Sometimes I turn it into a small interview of a few turns. But I find that the LLMs are somehow very good at reconstructing long incoherent rambles and often their echo of your own tangle of thoughts comes out quite a bit cleaner than what you started with. The result is that you improve the mind meld and have to correct things less from that point on.
It’s crazy the amount of UIs and APIs I have learned over the years from my work. Hubspot, Airtable, canvas lms, apps script, all of chat and Claude and what not APIs, amount of agent sdks, project management UIs, transcriptions ui and APis. The TERMINAL. So much more to go!
In a long time Siri has started to show good innovation and development. The intent to action has become sub ms for a lot of tasks finding, making events, setting timers reminders all of this has become direct intent to action when talk is done. The new Siri upgrade will be nice
I love making menu bar apps on my Mac for myself. My most recent one is where I just collect data on my app switches with duration spent on each window and Claude can talk to this data. It’s fun to know how i work and what i can improve!