Phoenix.new

Operator's take

Most vibe-coding tools get you a UI scaffold and then leave you to figure out the deployment yourself. Phoenix.new closes that gap: describe what you want, and the AI agent builds a full-stack application — database, backend, real-time frontend — and puts it live on Fly.io. The specialization in Phoenix LiveView is the bet here: you get real-time collaboration and live updates baked in by default, not bolted on. For a founder who needs a working prototype that doesn't fall apart the moment two users are on it simultaneously, that's a meaningful starting point rather than a toy.

The honest constraint is the opinionated stack. Phoenix.new builds Elixir/Phoenix apps, full stop. You're not picking React vs. Vue, you're not bringing your existing Node codebase, and you're not connecting this to a Bubble database. What you get is a coherent, production-capable runtime managed entirely through the browser — no local dev environment, no Dockerfile, no Git push required. That's a real advantage if you just want a thing that works; it's a wall if you need to integrate tightly with existing infrastructure or hand off to a developer who doesn't know Elixir.

Pricing is a flat $20/month subscription, which includes $20 of monthly usage credit — Fly.io staff posted the structure on the community forum at launch (June 2025). The catch is credit burn: early users report the included $20 runs down fast during active agentic building, and the per-month value has felt thin next to general-purpose tools at similar price points. Worth treating this as a prototyping and validation tool rather than a primary coding environment until you've calibrated how many sessions that credit actually buys you.

What it's good at

• Real-time apps out of the box — Phoenix LiveView handles live updates and multi-user collaboration without extra configuration; you get a genuinely interactive app, not a static page.
• Zero local setup — the entire development environment lives in the browser: database, server, preview, and deploy are all managed for you.
• AI agent writes and debugs — the agent doesn't just scaffold; it iterates, tests, and fixes errors as it builds, reducing the back-and-forth of a typical AI coding session.
• One-click deploy to Fly.io — finished app gets a public URL immediately; no separate hosting account to configure.
• Built by the framework's creator — Chris McCord (Phoenix framework author) built this at Fly.io, so the AI's understanding of Phoenix patterns isn't a bolt-on translation layer.

What it's not

• Not stack-flexible by default — the agent is tuned for Elixir + Phoenix + LiveView; other stacks (Rails, Node, Go, Svelte) are technically possible since those runtimes are installed in the VM, but you're outside the optimized path and results vary. If Elixir isn't your world, a general-purpose tool like Lovable or v0.dev fits better.
• Not a fit for complex integrations — early-stage tooling means apps requiring deep connections to existing SaaS APIs or databases will hit limits quickly.
• Not production infrastructure (yet) — treat output as high-quality prototypes; apps run on Fly.io infrastructure which is solid, but the phoenix.new agent layer is still maturing (file-rewrite behavior, validation loops).
• Not for teams without any technical oversight — someone needs to understand what they're shipping when it's time to maintain or extend the app beyond what the AI built.