programming has changed and its not going back. check out the IDE i'm building for a new way of programming: working with a team of ai agents
Discussion
Looks interesting. What language are you using for this IDE? I’d be curious to play around with it.
its a rust + egui native app. You can check out notedeck_dave crate in the notedeck repo to see the foundation of it
Now you need an agent to manage all the agents 😅
human in the loop is key or else it can unravel into chaos
For now... :)
I feel like this is not new, isn't this exactly what cursor allows you to do?
Isn't this just called the orchestration pattern or something?
Is it just the UX that you are focusing on here?
isnt cursor a single non-claude-code agent?
cursor is already outdated....
Fun fact. My advisor at MIT is doing research on population LLMs with his team running 50 million LLM agents at the same time. 😅
I have been trying to understand agentic setups recently, and afaik, according to LLMs ( 😁) cursor uses a special agent called "composer-1" as its orchestrator before handing off work to the real llms.
You can configure it to use claude btw if that was your question. It's just expensive af.
Lots of little tools run multiple agents. None of them are really good though. Android Studio. for instance, does it by opening new tabs for each agent. But again, it's not really that great of a UX. Most of the time I revert back to just one or two agents at a time.
These guys are doing a similar approach (multiple agents/roles) and are OSS.
yeah roles and workflows will be cool eventually. Im just trying to nail the key bindings + ide vibe atm
Can the agents interact? I guess they have filesystem access so they can leave messages in a text file or commits messages for other agents, but would be cool to create links between agents and let them talk through a local nostr relay
definitely have relay stuff planned, but keeping that secret until we launch
all the workflow features and plugins are also already a part of claude-code, this is really just a simple gui on top of multiple claude code instances. Each agent could be referencing issue trackers and things using mcp tools
Very interesting. Ai is killing so many jobs so this will be the only way to keep your head in the game
Super cool, Will! I should start a new career. This stuff is exciting. Let me know if you need an unpaid intern.
Isn't it like running Gemini CLI in multiple tmuz pans ..
I was doing this with claude but all you really get for grabbing your attention is desktop notifications
with this there is a priority queue jump list that you can jump through with a key binding
Notifications is a cool idea .. cuz if I am running CLI in separate windows then I need to switch to see the status ..
I think future is Molt Bot ish .... check out nostr:nprofile1qqsq743lut8a7xqvkyz9s6u4suehngxplh8mcvq6sryz2hen690s88gpz4mhxue69uhhyetvv9ujuerpd46hxtnfduhszythwden5te0dehhxarj9emkjmn99uq3uamnwvaz7tmjv4kxz7fwwpshgunfvd4h2mrjd93kstnrdakj7yfqegm 's PR .. cuz I suspect you would be using Damus to instruct your agents to write code , debug and deploy..
So sick! I’m telling you, man! Gamify it! Give those agents 3D avatars
haha might try it for the cool factor alone.
It would be cool if they were little units where you could see what they are doing based off their animation
That would be next level
I assume you would need to map them to tasks: researching, coding, waiting, etc and then have sprites or avatars that match those things
Kimi 2.5 adopted a similar approach:
Crazy! Of course everybody wants something like this! They might don't know it yet. Dev or not
