Show HN: P0 – Yes, AI can ship complex features into real codebases

arndt

Hey HN - Arndt here, one of the builders of p0. These always provoke some colorful discussion around here, so let me give some background. We've been building p0 because we kept hitting the same wall: AI coding tools are great at generating code from scratch, but can fall flat when shipping complex features into multi-repo codebases with real architecture, real standards, and real constraints. We'd get impressive results at first glance, then spend hours fixing the output to match our actual patterns. p0 bundles two things: a Mac desktop app and a purpose-built harness that treats feature development as a structured pipeline, not an open-ended chat. How it works: You start with a product spec (markdown) or an idea and end up with a set of PRs for all the repos that were touched. p0 runs through a 5-phase pipeline in isolated Git worktrees: 1. Import your spec, and/or brainstorm with AI to refine it, grounded in your codebase and standards 2. A specialized agent breaks it into phased tickets with technical contracts (acceptance criteria, architecture prescriptions, dependencies) 3. Engineering agents implement tickets in parallel while you watch a live ticket tree and agent activity in real-time... or grab a coffee 4. QA agents run through verification loops to enforce the contracts 5. Refine and create PRs in your repos Why not just use Claude Code Plan mode / [name your tool] We actually use Claude Code under the hood. What makes p0 different from Claude Code CLI / Conductor / etc. is our focus on shipping complex features autonomously, across all your existing repos. Spec first -- Puts humans in control of as much product and technical details as you can imagine, and we help you create/refine with agents grounded in your codebase. Contracts and QA loops -- We generate clear acceptance criteria and boundaries for each task. QA loops make sure they were adhered to. Ticketing -- The architect breaks every feature into phased tickets with dependency ordering. This isn't just a simple plan, it allows us to break complex problems into smaller, context-fitting tickets, and bring them back together into one cohesive implementation. Subagents -- Nothing fancy here, just a set of roles we've fine-tuned for months, so you don't have to start from scratch. Standards -- Typical coding agent behavior is to get a cursory (haha) understanding by reading code. But that clogs up context quickly, and rarely rises to the architecture understanding level. When you first launch p0, we'll help you generate a better AI-targeted documentation. Multi repo -- The whole harness is multi-repo aware. It maintains cross-repo context (imports, API contracts, shared types) and creates coordinated worktrees across all your repos in a single session. Local-first, team features through the cloud: All code stays on your machine on isolated worktrees. We do sync codebase documentation and workspace setup through our cloud so your teams can share those for convenience. And of course the prompts go to Anthropic's API. Limitations: - The spec-driven workflow has a learning curve. If you're used to the "chat away as you go" flow, the structured planning is a new thing to get used to. - macOS only right now, Linux/Windows are on the roadmap. - Works best for substantial features -- for small features, you're better off using Claude's plan mode. - Requires a Claude subscription or API key and works best with high limits / the 20x plan. Everything is finetuned for Claude 4.6 Opus right now. We plan on supporting other providers, but Claude is where the quality bar is. What we'd love for you to try: Is the spec-driven workflow helpful in building larger features? Did we miss anything? What integrations matter most? (We support GitHub, Gitlab, Linear right now) You can download p0 at https://www.bepurple.ai . Happy to answer any questions about the architecture or approach.

the_tli

We've started to be an early user in December and have since adopted it in a brown field codebase. I'd describe this as Lovable 2.0 or vibe coding 2.0. The p0 workflow allowed us to delegate medium to complex full features to Claude Code while staying in lane with our standards. It allows us to go from idea to fully working prototype and PR within half and hour to hour, most of the time fully hands off. PRs still need to be reviewed for production. However p0 allowed to drastically improve per engineer velocity, AI code quality and iterate faster with working prototypes and refinements. Compared to Claude Code directly, which we also use heavily, p0 keeps very strong coherence from user story, spec planning, architecting, engineering and QA - across many several agents and subagents. Breaking down the work into sequential and parallel task. With Claude Code alone this would be usually requiring lots of hand holding, or be be only partially focussed, rest lost in the woods. Also, we attempted to replicate some of p0 ideas with home grown software dev personas and workflows which fell apart. I think the strong point of p0 is that they really nailed the decomposition and software dev cycle with agents. Really recommend to try, and at the very minimum you get to make your codebase agent ready if you haven't already.

mareko

Interesting approach to put a lot of emphasis on the spec. Does this only work on existing codebases?

HrubyOnRails

I first tried p0 about a month ago. What stood out to me was the way the repo onboarding is designed. The harness puts you through a 15 minute Q&A to generate "standards" that it then keeps in sync with the codebase. IMO that is the best implementation of self-generating repo rules I have seen.

ngs-schlingel

Actually really promising! How does it differ from Claude Code? Massive opportunity right here!

franziloew

I started using P0 on our legacy codebase a couple of weeks ago and so far have shipped 3 features to production. Very helpful tool!

jasper_platz

Compared to Claude Code etc where do the time savings come from? Fewer PRs that need manual intervention? Higher quality code base to maintain later?

coscreen

Worked great for me. Think about integrating the /counselors skill so you can fan out complex questions to multiple hq models across Opus, Gemini, and Codex. We're using it to do extensive design/performance/SEO audits that benefit from multiple smart opinions.

cowling

Do y'all automatically feed information back into the spec and refine that over time?

niklasd

I tried it out and it worked really well for us – awesome tool! The spec-driven approach is a bit different then the usual back-and-forth with agents, and if you invest properly in writing specs it pays of.

isaackeitor

Standards generation is great. I really like that approach. It feels much better than relying on just a single CLAUDE.md or AGENTS.md. Does this work with Claude subscriptions like Max?

Franz23

I found myself talking to Claude in plan mode for hours before it ever ships a line of code.... and then still monitoring the execution all the time. Curious to give this a spin!

trufl-ai

seems to be pretty heavily tied in with claude? what if i would want to try out different models at some point?

jokineno

Subtraction is the new addition. As the cost of writing software drops, it’s so easy to just keep adding stuff. p0 takes kind of the opposite and refreshing approach. it forces you to focus on the fundamentals. What should we actually build? What’s truly essential and core? What can we ruthlessly cut? It directs you to think deeply instead of just vibing and shipping. Really appreciate that discipline.

shmoogy

Sounds interesting, but can't find pricing information anywhere ? Only: > We offer a subscription model per team. Token costs for AI models (Claude, etc.) are paid directly via your own Claude subscription or API keys. Reach out for specific pricing based on your team size. Is it "free" for single developer workflows ?

t3killer

Looks super interesting, and it's a great team.

ahoelzl

Great team behind it!

bernihackernews

insane team behind this, will be big

aburroughs

This is crazy cool. What does your roadmap look like?