Superconductor

Claude down? No problem. We have all the agents.

We ❤️ @PostHog

Meeting bots should SHIP CODE, not take notes. We just added a Meeting Assistant that listens for features and bug reports and automatically launches coding agents to implement them. By the end of the meeting, you have live previews of multiple new features!

@DanutPralea @dejavucoder We’ve been talking about it for months :)

@dejavucoder a shared claude code session you can hand off to someone else is genuinely the future of pair programming and nobody's talking about it

ramp's background agent has this btw. you can search on google ramp background agent blog

@dejavucoder You can use any coding agent multiplayer on Superconductor. Check it out!

GPT-5.4 is now live in Superconductor. And in our experience, it's a very good model! But find out for yourself: run codex with GPT-5.4 against Claude Code, Gemini, Amp, and Opencode on the same tasks, and see which one actually performs best for you!

GPT 5.4 in the Codex harness hit ALL-TIME HIGHS on our Rails benchmark. Both cheaper and better than GPT 5.2 and Opus/Sonnet models (in the Claude Code harness). You can test it yourself on your own codebase (whatever the tech stack) below.

@arafatkatze @AmpCode Already exists here at Superconductor. Can run Amp, Codex, Claude Code, Gemini CLI, Opencode. Browser based, native mobile app, as many agents as you want.

I am calling it now that the @ampcode team will make an app-like overview interface where tons of chats can run in parallel, as every Agentic conversation is bottlenecked by inference so users will be able to run many different chats in parallel for different PRs in different Repos each operating in a separate worktree or ideally in separate cloud runners for proper separation of concerns. The purpose of the "dead" IDE would then just be the basic last resort check to "see with your own eyes" but ideally you would just look at the generated PR and iterate a bit on each. In this world, the IDE is dead because it doesn't really matter which one you use as you can use just about any and the only real bottleneck is the humans ability to interpret understand and most importantly context switch between different conversations. I will keep this post here and when they come out with a release I will cross post it there.

winning dev loop: mid-PR review, your tool ships a feature fast enough to help with that same PR review and the whole loop happens from your phone

@BHolmesDev Yep! We realized worktrees were a temporary solution last summer and built the best way to run them in the cloud. They can send you screenshots and videos, and you can control them with a native app, including being able to interact with your live running app yourself.

Every day I'm more convinced that worktrees are a band-aid solution. Putting agents in cloud runners lets you *actually* close the laptop, and gives agents a space to check their work with sandboxed screenshotting / e2e testing. Y'all experiment with this yet? I'm still early

@ryanvogel Try out @RunloopDev

going to benchmark sandboxes i have - cloudflare - vercel - daytona - e2b.dev - exe.dev - sprites anyone else i should try

@ericzakariasson We’ve been doing this for months — 95% of our work happens on remote sandboxes. With a native mobile app.

the next version of cursor cloud agent is actually one of the few things that's changed how i work and approach software development. majority of my work now happens on a remote machine, and i can monitor & intercept when needed, from anywhere (web, phone, desktop). excited to share more soon!

Claude Sonnet 4.6 is live in Superconductor! Benchmark results forthcoming -- let's see how it will do against Opus 4.6 and GPT 5.3 Codex on our Rails codebase.

@steveklbnf @ssh_exe_dev @conductor_build @superset_sh No full corporate subscription required here. Just fast forked VMs.

i think this is the future conceptually, but i don't want to go full corporate gitpod subscription i think the future looks like @ssh_exe_dev (fast instant forked VMs) with a hosted @conductor_build or @superset_sh like experience

@jeff_weinstein @stripe Every company has 100 of these "we were busy on other things" tickets rotting on the backlog. Minions are cool but internal to Stripe. superconductor.com is for the rest of us.

[friday fun!] for 15+ years, @stripe docs didn't have dark mode. (we were busy on other things.) yesterday, we asked our minions to try it. they worked for an hour and, honestly, did a pretty good job! try `cmd`+`shift`+`d` if you see a bug send to jweinstein+minion@stripe.com

@stripe Congrats! Next step: give every agent its own dev environment with a live preview of the running app. Seeing what each agent actually built before touching the diff is a HUGE unlock.

Minions are our homegrown coding agents. Over a thousand pull requests merged each week at Stripe are completely minion-produced, and while they’re human-reviewed, they contain no human-written code.

Time to docsmaxx!

@stevekaliski This is the way. We've been building the same workflow for teams that can't spin up a dedicated infra team to make it happen.

At Stripe we have a tool called "minions" -- it lets us kick off async agents built right in our dev environment to one-shot bugs, features, and more e2e. I have team, project, and personal channels dedicated just to working with minions. I like to think of it as a new type of pair programming -- "pair prompting." Read more --> stripe.dev/blog/minions-s…

@ericclemmons 100% agree. we replaced worktrees last year so our sessions could be accessible on mobile and multiplayer.

I think worktrees are a short-term solution to what should be remote sandboxes.

@martin_casado Also, in the one-shot mindset, run multiple agents on the same task, and flip through to find the one that’s closest to what you want. Don’t try to course correct an agent that’s not on track.

Couple things: 1) every time an agent does something you don’t like, ask it why it did it, and how it would change your Claude/agents.md to prevent that in the future. Clean up the suggestion and commit it. 2) try and get the agent to one shot whatever you’re doing, and then something like our guided review to review it faster, and leave comments on the diffs as you review to fix it up. Much faster than the normal IDE based flow. Then do 1) x.com/sergeykarayev/…

I'd love tips on how you all handle code complexity when AI coding the same project for months. I end up spending like 7-80%% of the time just cleaning up and refactoring and deleting things. There has to be a better way.