Permaship AI

PermaShip is now live. A persistent engineering team that runs continuously on your codebase. From request to verified PR in 22 minutes. It's not a vibe-code tool. It's not an IDE plugin. It's engineering-grade confidence with approval gates you control.

What happens when the tools you used to prove your worth become obsolete? When building stops being the source of your identity and proof of existence? #AI #AIAgents #SoftwareDevelopment #IdentityCrisis #FutureOfWork #AIImpact

We open sourced Nexus Command. An AI team that runs on your codebase — proactively finds work, proposes it, reviews it, and executes it through Claude Code, Codex, or whatever you use. MIT licensed. One command to run it: npx nexus-command github.com/PermaShipAI/ne… #opensource #aiagents #devtools #buildinpublic

Copilot and Cursor are fast hands. You still have to point them. The next wave of engineering tools doesn't wait for you to ask. It finds what's worth doing and makes its case. Big difference between a tool that's fast when you use it and a system that runs when you don't.

The real bottleneck in AI agents isn't coding, it's arbitration. Complex software changes need aligned priorities. Structured disagreement surfaces trade-offs, leading to better decisions. #AIAgents #SoftwareDevelopment #ExecutiveFunction #MachineLearning #TechInnovation #FutureOfAI #AIStrategy #Coding

Every sprint planning meeting, the same low-priority work gets pushed again. Security fixes, test gaps, dependency updates. Nobody has capacity. It compounds. The problem isn't that your team is slow. The problem is that toil never wins the prioritization fight. #AI #Agentic

Download "The End of Sprint Planning: A Manifesto for Continuous Engineering" to read more. buff.ly/4wF2CtC #AI #SprintPlanning

Our full manifesto covers the future without sprints: • The five pillars of continuous engineering • 90-day transition framework • Common failure modes to avoid • What engineering orgs look like in 2028 • Why early adopters get 12-24 month advantage

The two-week sprint isn't a law of physics. It's a coordination artifact built for human constraints. Agents don't have those constraints. 🧵

Software cost is trending toward zero, and TAM barriers are collapsing. Niche markets that were too small to justify building for are suddenly viable. The moat isn't that you built the software anymore. It's distribution, user acquisition, and understanding the customer well enough to reach them inexpensively. #AI #Claude

@bygregorr @hasantoxr that's why we built permaship to handle the triage and verification. it runs the full CI and writes reproduction tests before it even pings you. it filters out the noise so you aren't just a full-time reviewer.

@hasantoxr Running 30 agents in parallel sounds wild until you realize you now spend 3x longer reviewing 30 PRs than you saved generating them.

🚨 BREAKING: Composio just open sourced a tool that lets you run 30 AI coding agents in parallel on your codebase. It's called Agent Orchestrator and it's sitting at 4.1k stars for a reason. One command spawns an agent. It gets its own git worktree, its own branch, its own PR. CI fails? Agent fixes it. Reviewer leaves comments? Agent addresses them. Here's what it actually handles: → Spawns parallel Claude Code, Codex, or Aider agents → Isolated git worktrees so agents never step on each other → Auto-fixes CI failures by sending logs back to the agent → Auto-addresses code review comments without you touching anything → Real-time dashboard at localhost :3000 to supervise everything → Notifies you only when human judgment is actually needed One command: ao spawn my-project 123 That's it. Without this, you're manually creating branches, babysitting agents, reading CI logs, forwarding review comments, and tracking which PRs are ready. With this, you spawn and walk away. This is the missing coordination layer every team running AI agents in production needs. 100% Open Source. MIT License. Built by Composio. (Link in comments)

@glorymaxing @BrendanFalk we see this all the time. one-shotting is a myth for anything non-trivial. permaship uses a loop of linting and test verification because the first prompt is usually wrong. testing how a dev handles that 'fix-it' loop is the real signal.

@BrendanFalk This is actually super insightful. I’ve been struggling so hard with conducting interviews in this day and age, everyone wants to one shot with Claude.

I believe we've found the best AI-native coding interview We call it the “Composer 1 interview” Candidates get 1 hour to build a real, medium-sized project live The only constraint: they have to use Cursor’s Composer 1 model

@lrnsdesign long sessions are just a slow way to find out you forgot a semicolon. 20 minutes in and i'm usually just fighting a linter. i'd rather just toss the ticket to permaship and go get lunch.

This is why long cursor sessions have felt more stable lately. 1/5 of the tokens, better accuracy. Less context drift mid-task. Good time to still be on cursor 😃😃

@anthonysheww @mitchellh @turborepo terminal output is a nightmare to get right. we had to build custom sandboxes at permaship just so our agents could handle ci logs reliably. ghostty core for turborepo would be a huge win for stability.

@mitchellh Hey, Mitchell, will there be Rust bindings (or another blessed path)? We have a TUI for @turborepo that I think could be more stable, and I'd love to lean on libghostty for a solid core. Happy to ask/contribute in a more appropriate forum if need be.

More libghostty C progress: encoding APIs for mouse reports, focus events, and mode reports. Mode read/write on terminals. The current goal is to expose all major event/response formats as high level APIs. The core idea is to minimize the glue between UI framework events (mouse events, key inputs, focus, etc.) and libghostty to provide accurate terminal emulation.

@luongnv89 @lawrencecchen having 40 active sessions is just a slow-motion crash. cmux is basically just a better rearview mirror for the pileup.

@lawrencecchen haha, 2 my pain points nowadays: - context lost (too many open sessions) - lost attention (don't know which one need my attention) now testing cmux for the cursed. so far so good, my become my default terminal soon. thanks

My favorite cmux feature is Cmd+Shift+U. I have 17 workspaces open right now, each running an agent. I used to click through tabs and macOS system-wide notifications to figure out what completed. But with cmux, I can just "Cmd+Shift+U" Cmd+Shift+U jumps to the newest unread notification. In practice that means the last agent that finished. It switches to the right workspace, focuses the exact pane, flashes it so you see where to look, and marks it read. If the notification came from another window, that window comes forward.

@1258632 @ctatedev telegram auth is its own brand of headache. one minute you're doing oidc, the next you're fighting botfather settings because a redirect uri is slightly off. standard examples rarely survive the encounter. good luck with the widget logic.

@ctatedev does it work for telegram oicd? they require url when implementing auth using telegram

New skill: OAuth + Portless Many OAuth providers reject .localhost and .test subdomains Portless fixes this with --tld to give your app a valid TLD locally Give your agent this skill: npx skills add vercel-labs/portless --skill oauth

@Lat3ntG3nius @minchoi vector search is just vibes. once you get past a couple hundred files it's just a guessing game. glad to see people moving toward graphs.

@minchoi Knowledge graph over flat file retrieval is the right call for large codebases. Call graphs, dependency trees, and module ownership are not well captured by embedding similarity alone. The MCP bridge makes this actually usable by agents mid-task rather than as a one-shot lookup.

Ok this is insane. GitNexus turns your codebase into a knowledge graph that Claude Code, Cursor + other AI coding tools can use through MCP. Helps your AI coding assistant understand your entire repo. Public repo + demo in reply 👇

@lemir_rodriguez @rauchg we're doing this with permaship. a slack message or linear ticket starts the loop, but the agent works in a sandbox with its own ci and audit logs. the chat is just the trigger for the orchestration.

@rauchg Chat is becoming the default entry point, not the whole product. The winners probably pair conversational UX with hard system affordances underneath: workflows, permissions, state, audit logs.

Every month I periodically see the recycled take that “something better than chat” is coming for AI. That chat is temporary. In fact, I predict the opposite. More of our work and life will happen through chat and voice interfaces of increasingly intelligent agents. 🦞 OpenClaw is the lobster in the coal mine of this. It’s true that as humans we consume with all the senses. When we use our internal company agent at @vercel, it can answer any question in English, but also plot data and richly visualize it. All in the chat medium. You start with any question, and receive periodic reports, much like an enterprise claw, on @slackhq. If the chat is not enough, you’re one click away from more depth and refinement on a web page. I don’t believe web pages are going anywhere. Many will evolve to accept natural language, which is the lingua franca of AI, and stream both text and complex data. We call this Generative UI (eg: json-render.dev). And web pages will also be crucial in enriching the conversations that happen elsewhere, whether you’re chatting with your agent on WhatsApp or Slack. We’re building chat-sdk.dev to make this universal interface as easy to deploy as possible.