@patrickc how do you resolve context conflicts when two people edit files mid-inference without resetting the whole run

- Ability to manage a set of input files (Markdown or similar), plus other general-purpose context. - Some concept of compiled outputs/inference results (which ideally can be shared externally).

Both of those yes, but agent native: input and output artifacts oriented toward compounding engineering workflows. So output is just reusable context as standards.

https://plannotator.ai/workspaces/

But for better or worse, zero intent to do any orchestration; integration instead.

@patrickc RIP your DMs

@patrickc We’re building this at http://runtm.com, we’re in the current YC batch :)

- templates (context, guardrails, coding agents) snapshotted. - 80% of our sessions are multiplayer - working with a fintech unicorn and other scaleups :)

@patrickc It’s Paperclip

@patrickc You might like something you build and run in codex like this:

@patrickc I'm working on dynamic workflows + typed artifacts @overmindai. coincidently, I saw this excellent manuscript that formalizes some of the ideas I've been thinking about beautifully

@patrickc In my (anecdotal, but enterprise deployed!) experience, this ache is a product grammar / customer mental modeI problem. I landed on something between Slack, Confluence and Figma, built around "work lanes" (a mental parallel to a channel/workspace/folder).

@patrickc Stilla is multiplayer, real-time collaborative, individual level permissioned and version controlled from the ground up across everything (chats, skills, agent workflows, wiki etc)

What are you looking to do?

@patrickc I think you'll enjoy what @sgrove has been working on!

@patrickc i built my own system with similar properties.

just used it to one shot a pokemon ROM binary to WASM static recompilation.

a task that should have taken a team of 10x devs a year to complete, my agent system did it autonomously in ~4 hours

this is part of @inline_chat's vision as the interface + realtime interactions + multiplayer + context/artifacts/storage + share externally. it works much better than slack/discord/etc because threads are the primitive, can be linked/referenced as context, individually shared with external parties, and infinitely scalable while multiplayer with no conflict/bloat as you only see what you need. the agents/workflows building on top of it won't be built-in though. they'll interact via our apis.

@patrickc we're literally building this at http://AgentSwarms.fyi

visual sandbox where you can wire up agents, manage context/files, and iterate on workflows without setup. outputs are shareable and everything runs in browser

would love to show you what we've got

Not sure it’s an exact fit but been thinking along these lines with https://schematics.run . The idea is you could define schemas/tools for different files in a file system. Maybe you want *.pdf to generate screenshots then use those to generate markdown and then use the markdown to generate json in a certain shape, track dependencies and relations across all the files etc

@amplituhedron @patrickc @overmindai Glad to hear!

@patrickc I moved to a canvas-based orchastrator called STAX IDE. MY work is a lot of human-in-the-loop and changing layouts

Building this.

File-canonical, agent-agnostic, build artifacts that stick around and feed each other — the "all this stuff I want to compute over iteratively" problem is the entire thesis.

Real-time collab is the one piece I've deliberately held back on for now, the GitHub-as-AppStore seems like a more appealing route.

Happy to show more

@amasad @patrickc It’s been a game changer for me as a solo non-tech founder to have everything needed in one place essentially! @amasad Thank you for bringing your vision to life so that I can do the same with mine!

@patrickc Real time collab is the tough part here - can build this with @instant_db cc @stopachka

@patrickc What we’re building @obvious