That’s exactly my process. I created an adversarial review gauntlet as a skill and workflow combo that has 3 rounds of subagents review the brief and implementation with Opus first, then external review by GPT at the end of each 3 rounds. The 3 rounds repeat until GPT and Opus are in alignment before moving on to the next phase.
@doodlestein We built worktree management, draft PRs, and controlling the merge flow directly into @CoordHQ for this reason. You don’t have to think about it. It just works cleanly behind the scenes. Review branches locally as needed. Your agents work on the same repo without issue.
@petergyang For #1 check out @CoordHQ. It lets you run your agents on any computer you want while the orchestration is in the cloud so you can manage them from anywhere. It also has scheduled jobs to replace cron. Coord support running Codex, Claude, and Gemini.
AI doesn't replace creators. It replaces the slow part of getting their ideas out.
If you've been making your living executing other people's ideas… you might want to stop worrying about AI and start working on having your own.
@mattpocockuk I noticed it highlighting “workflow” so now I have to consciously type “work flow” instead. It makes no sense to me why this wouldn’t be a / command.
Someone told me this weekend that humans will be "out of the loop" by 2027.
Doubtful.
The future I see isn't agents replacing people. It's people and agents working together.
Dev speed doesn't change the process. How to collaborate efficiently has always been the hard part.
@burkeholland Just because everyone has a camera doesn’t mean they’re all photographers.
The ability to create has been removed, but what you create matters more.
@rauchg Cloud based orchestration for local agents focusing on team based agentic development. Mostly Claude with a dash of Codex. https://t.co/NBvbXLtGUa
@burkeholland@codemullins Why is level of effort the bar? This feels like the discussion I have regularly about clients wanting us to provide actuals for our hrs on creative projects. You’re paying for the result, not how long it took me to do it.
Time and effort does not equal value.
@swyx I’m building a new platform now. Spent 4 days and 26 rounds with multiple subagents between Codex and Claude reviewing each others work just to solidify the api auth structure and foundation.
People that think they’re one shotting are getting a demo, not an app.
Everyone walked into Code with Claude expecting a bigger model…
What we got was managed agents, advisor mode, remote sessions, CI auto-fix.
That's not a holding pattern. That's an admission.
The bottleneck isn't the model anymore. It's the orchestration.
@NickADobos Riddle me this... so I can run multiple agents simultaneously in a UI as long as it's their UI, but if I do the same exact thing using my own UI I'm taxed for it?
How does that make sense? They already let you do it, but their UX sucks. Their needs to be an ADE carveout.
@Sentdex The issue is that they gate you into using their desktop app for any level of managed orchestration. The desktop app doesn't cost money, claude does. So why charge me more for wanting a different UX for the same claude functionality?
If using Claude Code on the desktop app for multi-agent sessions retains subscription pricing, what is the reason you don’t want to use it?
For me the UI is ok, but I need a shared backlog, and kanban agent sessions with visibility across my dev team.
@burkeholland I’m thinking we need verified phone numbers now. I can’t even get real calls from businesses anymore. My bank tried to call me… went into a list of 151 voicemails. Bank’s support said they tried to call me. I said there was no way I would know. They agreed it’s a real problem.