Show HN: Agentic Ship – open-source Lovable alternative that runs on your agent

What happened

I'll write the post body from the Show HN facts only, then check word count and strip any markdown.The draft is 906 words, just over the cap. I'll trim it into the 600–900 range without adding facts.Two words over. I'll cut a few phrases and keep the same structure.Show HN introduced Agentic Ship as an open-source harness for shipping a fullstack application the agentic way. The project is framed as a Lovable alternative that runs on the agent the builder already uses, rather than as another hosted app-generation product. The stated cost model is narrow: the user pays for an existing AI subscription such as Claude, Codex, or Cursor, plus a domain. That is the commercial claim in the announcement. Everything else in the pitch is about what the harness coordinates once that agent and that domain are in place.

The mechanics described are those of a coordinator, not a single-vendor runtime. Agentic Ship is said to start by automatically connecting to the providers a fullstack app needs, with Convex and Netlify named as examples. From there it is described as building the frontend, the backend, payments, and tests, then carrying the same work through to production deployment. The announcement insists that this sequence is coordinated with careful engineering rather than left as a loose prompt chain. In that model the agent is the worker and the harness is the shipping procedure: provider wiring, surface generation, payment and test coverage, then a live deploy. The user does not buy a new builder seat. They point their existing agent at a process that already knows which integrations to open and which stages to finish.

The technical detail

That split matters for engineers who already pay for Claude, Codex, or Cursor and who already treat those tools as the place where code is written. A hosted AI builder typically sells generation, preview, and deploy as one billable surface. Agentic Ship instead treats generation as a cost the user has already accepted, and treats shipping as the missing layer: connect Convex, connect Netlify, produce frontend and backend, add payment, add testing, and leave the app on a domain the user owns. Builders who want the agent in their own loop, rather than inside a third-party canvas, get a path that does not ask them to abandon the subscription they already use. The open-source framing also means the shipping procedure itself can be inspected, forked, or constrained, which is a different control surface than a closed builder that only exposes a chat box and a preview pane.

Why it matters for builders

The competitive claim is explicit. Agentic Ship is positioned as a replacement for AI builders such as Lovable, Bolt.new, and Replit. Those products are the named contrast class: hosted environments that generate and run apps for you. Agentic Ship answers that class with an open-source harness that rides on Claude, Codex, or Cursor and with automatic hooks into providers such as Convex and Netlify. The substitution being offered is not another AI that writes a frontend. It is keep your agent, keep paying only that subscription and a domain, and let a shipping harness drive frontend, backend, payment, testing, and production. Anyone evaluating the Show HN against those three names should score it on that substitution, not on whether it can emit a landing page.

The practical test is whether the harness actually closes the loop it describes. Watch whether automatic provider connection to Convex, Netlify, and the unnamed rest of that list is reliable enough that a builder does not fall back to hand-wiring each service. Watch whether payment and testing are first-class stages or afterthoughts that the agent sometimes skips. Watch whether production deployment is a finished Netlify-style publish onto the user’s domain, or a preview that still needs a human to finish the last mile. And watch whether the same agent subscription that writes the code is sufficient in practice, or whether the harness quietly depends on extra accounts beyond the AI plan and the domain that the announcement names.

Market and competitive context

Several open questions sit on top of that test. Replaces Lovable, Bolt.new, and Replit is a product claim, not a measured result; those three products are not equivalent to one another, and a harness that runs on your agent will not automatically match their hosted previews, collaboration, or runtime isolation. Well engineering is asserted without a public description of the orchestration contract: what the harness requires of the agent, what it retries, and what it refuses to mark as shipped. Automatic connection to Convex and Netlify also implies a preferred backend and host; it is unclear how far the path bends when a team wants a different database or a different production target. Because the user only pays the AI subscription and the domain in the stated model, any required Convex, Netlify, or payment-provider billing sits outside that sentence and should be treated as an unstated dependency until the project shows otherwise.

What to watch next

Related prior art is the category the announcement is already arguing against. Lovable, Bolt.new, and Replit are the hosted AI-builder line: they keep the agent, the preview, and often the deploy inside one product. Agentic Ship is the inverse shape: an open-source shipping harness that assumes Claude, Codex, or Cursor already exist, then tries to own the path from provider connect through frontend, backend, payment, testing, and production. That is a recognizable split in agentic tooling between canvases that generate apps and scaffolds that make an existing agent finish a release. The Show HN is worth reading as an instance of the second pattern, judged by whether the harness can take an agent from an empty idea to a paid, tested, domain-hosted fullstack app without reintroducing the builder tax it says it removes.