Hosting for AI-Built Apps: Where to Ship Agent-Generated Code

Last Tuesday I spent 47 minutes prompting Claude to build a full-stack invoicing app. It generated a React frontend, a Postgres schema, three edge functions for Stripe webhooks, and a transactional email template for payment confirmations. Then I spent two hours trying to deploy it somewhere that wouldn't bill me $200 the first time a background task ran longer than expected. If you're looking for hosting for AI-built apps, you've probably noticed most options are static-only or surprise-metered. This guide is what I wish I'd had that Tuesday: a straight answer on where to ship agent-generated code that actually needs a database, server functions, and a URL your client can visit next month.

Table of Contents

• Why "One-Click Deploy" Usually Means "One-Week Headache"
• What Your AI Agent Actually Needs From a Host
• The Hosting Landscape for AI-Built Apps in 2026
• How to Deploy an AI-Built App: Concrete Steps
• The Hidden Costs Nobody Talks About
• Real Talk: When to Use What
• The Bottom Line

Why "One-Click Deploy" Usually Means "One-Week Headache"

The marketing is everywhere. "Deploy with one click." "Ship your AI app in seconds." What they don't say is that the URL expires in 24 hours, the database is a JSON file on your laptop, and the moment your app does anything interesting, the free tier evaporates.

Most AI agent hosting platforms give you a preview URL, not a production URL. That's fine for showing a teammate what you built during lunch. It's useless when a client expects their customers to visit the site tomorrow. Preview links die. Permanent URLs don't.

Then there's the static-only problem. Netlify and Vercel's free tiers are excellent at serving HTML, CSS, and JavaScript. But your AI agent didn't write a static site. It wrote server-side logic: database queries, API routes, form handlers, authentication flows. Static hosts can't run that code without workarounds. You end up refactoring what the agent built, which defeats the purpose of having the agent build it in the first place.

Metered billing is the silent killer nobody warns you about. Your AI agent doesn't know what things cost. It will happily write a while loop that polls a database every 100 milliseconds. On a metered host, that's a $300 surprise by morning. The agent can write the code, but it can't negotiate your hosting bill. That's on you.

The real cost isn't the build. It's the hosting. Free tiers vanish the moment your app needs a database or server-side logic. And by then, you've already built the thing. You're stuck either paying whatever they ask or rewriting for a different platform.

What Your AI Agent Actually Needs From a Host

Your agent doesn't care about brand names or venture funding announcements. It needs specific infrastructure. If the host doesn't provide it, you'll spend more time patching deployment gaps than you spent prompting the agent.

A Permanent URL That Doesn't Expire

Preview URLs are for iteration. Client handoff needs a real domain. The host should support custom domains with automatic DNS configuration, preferably through Cloudflare or a similar provider that handles propagation in under a minute. SSL certificates should be included, not an upsell. If you have to think about HTTPS, the platform already failed.

Real Backend, Not Just Static Files

Your agent built a SaaS app. That means a database. SQLite is fine for local development but it won't survive a deploy. You need something like Postgres or Cloudflare D1, provisioned per site, not shared across projects in ways that create security headaches.

Server or edge functions are non-negotiable. Your app has API routes, webhook handlers, authentication endpoints. Those need to run somewhere. If the host only serves files from a CDN, you're looking at the wrong category entirely.

Transactional email is another thing agents generate constantly. Password resets, welcome messages, invoice notifications. Your host should handle the plumbing or at least not block the SMTP calls your agent already wrote.

Predictable Pricing You Can Bill Through

Metered hosting means you can't tell a client what their monthly cost will be. Flat-rate pricing lets you charge a fixed fee without margin risk. You quote the client $50 a month, you pay $24, you keep the difference. That math only works when the host's price is predictable.

Ownership transfer matters more than most builders realize. When the project is done, you need to hand the site, domain, database, and functions to the client cleanly. Atomic transfer means one operation moves everything. If the platform doesn't support that, you're stuck as the middleman forever, forwarding bills and resetting passwords.

The Hosting Landscape for AI-Built Apps in 2026

The market splits into four categories. Each has a use case, but only one was built for what AI agents actually produce.

The Static-Only Trap

Netlify, Vercel, and Cloudflare Pages dominate the JAMstack world. They're excellent at what they do: serving static assets fast, rebuilding on git push, handling form submissions through add-ons. But your AI agent probably wrote server code. These platforms offer serverless functions as a workaround, but that adds complexity your agent didn't account for. Function invocation spikes generate surprise bills. Build-minute caps punish the rapid iteration that AI-assisted development encourages.

These are best for marketing sites, landing pages, and static portfolios. If your agent wrote a single HTML file with some CSS, deploy here and call it done. If it wrote a backend, keep looking.

The Platform Lock-In Problem

Firebase now offers "agent skills" that help AI tools understand Firebase-specific APIs. That's clever, but it also means your agent generates code that only runs on Firebase. Supabase has a similar story: great developer experience, Postgres under the hood, but you're building inside their ecosystem. Replit pitches "no coding required" and one-click hosting, which sounds perfect until you try to export your app and discover it depends on Replit's runtime, Replit's database, and Replit's authentication layer.

Migration out of these platforms is painful. Proprietary data formats, vendor-specific APIs, and configuration that doesn't translate anywhere else. These are best for prototyping, internal tools, and learning. If you're shipping client work, the lock-in risk is real.

The Flat-Rate Alternative

Shiply takes a different approach. One API call from your AI agent publishes any app, static or SSR, to a live URL with a per-site SQL database. No usage meter. Pricing is flat: free, $8, $24, or $49 per month depending on what you need. What you see is what you pay.

The platform includes edge functions through Workers Lite, real databases via Cloudflare D1 and Neon Postgres, transactional email, and a built-in inbox for form submissions. Custom domains auto-configure through Cloudflare DNS. When the project is done, atomic site ownership transfer moves everything to the client's account, and Stripe Connect handles the payout. You hand off the site, get paid, and walk away.

This category is built for freelancers, solo builders, and anyone shipping client work. The pricing is predictable enough to bill through, and the backend is real enough to run what your agent generates.

The Enterprise Overkill

AWS, GCP, and Azure can deploy anything. They also require DevOps knowledge your AI agent doesn't have. Cost estimation on these platforms is practically a full-time job. The services your app needs are scattered across a dozen consoles with opaque pricing models. Surprise bills are the norm, not the exception.

These are best for teams with dedicated infrastructure engineers and high-traffic production apps where the complexity is justified. For a solo builder shipping a client project, they're overkill that creates more problems than they solve.

How to Deploy an AI-Built App: Concrete Steps

The order of operations matters more than most tutorials admit. Do this wrong and you'll refactor for hours.

Step 1: Choose Your Host Before Your Agent Writes Code

Tell your AI agent the target platform in the first prompt. This is the single biggest time-saver in the entire workflow. Agents generate different code for different environments. If you build first and choose a host later, you'll spend hours adapting database connection strings, function signatures, and deployment configs.

A good prompt sounds like this: "Build a full-stack todo app with a Postgres database, edge functions for API routes, and transactional email. Target: Shiply flat-rate hosting." The agent now knows what infrastructure is available and writes compatible code from the start.

Step 2: One-Call Deployment From Your Agent

Most flat-rate hosts accept deployment via API call or CLI command. For Shiply, your agent makes a single HTTP POST with your build output. No account is needed to start. You deploy first, see it live, then decide on pricing later. Custom domains auto-configure through Cloudflare DNS in under 60 seconds. The deployment workflow is documented in the quick-start guide if your agent needs the specifics.

Step 3: Verify the Backend Actually Works

AI agents hallucinate API endpoints. It's not a question of if, it's which ones. Before showing the client, test every database connection, every edge function endpoint, every form submission. Send a test email. Trigger a webhook. Confirm the data actually persists.

While you're verifying, use a hosting bill calculator to compare what you'd pay on metered versus flat-rate hosting. The difference is often staggering, especially for apps with background tasks or polling logic. The numbers don't lie, and they'll justify your platform choice to any skeptical client.

Step 4: Transfer Ownership and Get Paid

When the app works and the client approves, initiate the atomic site transfer. This moves the site, domain, database, and functions to the client's account in one operation. Stripe Connect handles the payout. The client pays, you receive funds immediately. No ongoing hosting management, no surprise bills forwarded to you six months later.

This step is where the freelance business model actually works. You built the thing. You deployed it. You got paid. Now it's their problem. That's the whole point.

The Hidden Costs Nobody Talks About

Database egress fees are the worst offender. Some hosts charge you to get your own data out. If you ever need to migrate, that bill can be substantial. Check the egress policy before you commit.

Function invocation limits hide behind the word "unlimited." What it usually means is "unlimited until we throttle you." Read the fine print. If your app polls an endpoint or processes webhooks at scale, throttling will break it.

Build minutes are another trap for AI-assisted development. Agents regenerate code constantly. Every iteration triggers a rebuild. If your host caps build minutes, you'll hit the limit fast and either pay overages or wait until next month.

Bandwidth overages turn viral moments into financial headaches. A client's post hits the front page of Hacker News and suddenly you owe $400. Flat-rate hosts cap the cost. Metered hosts don't.

Support tiers on free plans often mean forum-only help. When your client's site is down at 11 PM on a Saturday, "check the community forums" is not an acceptable answer. Know what support you're actually buying.

Real Talk: When to Use What

For prototyping and learning, Replit or the Firebase free tier work fine. You're not shipping to clients yet. The lock-in doesn't matter because you're not committed.

For client work that's purely static, Netlify or Cloudflare Pages with a flat-rate plan are solid choices. Your agent probably didn't write a static site, but if it did, these platforms are excellent.

For client work that's full-stack, which is most of what AI agents generate in 2026, you need a flat-rate host with a real database and edge functions included. Shiply fits this category. The pricing is predictable, the backend is real, and the ownership transfer means you can actually close the project.

For high-traffic production apps with dedicated infrastructure engineers, AWS or GCP make sense. The complexity is justified by the scale, and you have someone whose job is managing it.

For internal tools, Supabase or Firebase work well. Vendor lock-in matters less when you're the only user and the alternative is a spreadsheet.

The Bottom Line

Most hosting for AI-built apps is either static-only or surprise-metered. If your agent built a real app with a backend, you need a permanent URL, a database, edge functions, and flat pricing you can bill through. Shiply.now does exactly that: one call from your agent, live in seconds, no meter, atomic transfer when you're done. The right choice depends on your workflow. Pick the host that matches how you actually ship, not the one with the best landing page.