9 min readBy Matt Delgado
AI Website Builders for Agencies: Lovable, v0, Bolt, and When to Hand-Code
Where Lovable, v0, and Bolt genuinely belong in an agency's delivery process, what a team seat actually costs, and the four edges where hand-coding has to take over.
Yes. An agency can build a real client prototype in Lovable, v0, or Bolt in an afternoon. Each one lets the code leave the platform. None of the three produces a finished site on its own. The client-specific wiring is still a build step. Each vendor also gates ownership differently. Here is where each tool belongs in a delivery process, what a team seat costs, and where hand-coding has to start.
What each tool is actually built to do
Three different shapes of output, from three different starting prices:
| Tool | Starts at | What it ships |
|---|---|---|
| Lovable | $25 a month, Pro, for 100 credits | A React app wired to a live database, described in one chat conversation. |
| v0 | $0 a month, Free, with $5 of monthly credits included | React and Next.js components and pages, built by the team behind Vercel and Next.js. |
| Bolt | $0 a month, Free, with 1M tokens a month included | A full front end and back end, running in the browser, one click from a live deploy. |
The buyer-facing comparison of these same three tools, Lovable vs. v0 vs. bolt.new, compared, breaks down which one to try first for a business building its own site. This post assumes the reader is building somebody else’s.
Where a builder genuinely earns a place in the process
Three moments where a fast, disposable build beats a careful one:
- The discovery-call mockup. A client describes their business in plain English and sees a real, on-brand page before the call ends. That single artifact does more for a proposal than a written scope ever will.
- The internal spike. Testing whether a layout, a flow, or a data model even works is exactly what a throwaway build is for. Do it before committing a developer’s week to the real one.
- The one-page microsite with no moving parts. An event page. A landing page for a single campaign. Anything with no form logic tied to a client’s CRM can ship from the builder with minimal follow-up work.
Where the tool stops and hand-coding has to start
The same four edges show up on every real client project, whichever tool built the first draft:
- A backend integration specific to the client. A booking system tied to their existing calendar. An IDX feed. A payment flow wired to their merchant account. A builder can scaffold the shape of this. It does not wire the client’s actual accounts into it.
- A regulated industry. Real estate, medical, and financial sites carry disclosure and advertising rules a generic build has no reason to know about, and a generated page will not know to ask.
- Anything multi-tenant. A site serving more than one business, more than one location, or more than one user role needs a data model designed for that from the start, not bolted on after the fact.
- Anything the client depends on for years, not weeks. Someone has to notice when the platform changes its terms, a dependency breaks, or the hosting bill changes. A generated site has no one watching it by default.
What “export the code” actually means, tier by tier
The three vendors gate ownership at different points, and the free tier is where the differences show up most:
| Tool | What the free tier withholds |
|---|---|
| Lovable | Lovable’s own documentation lists Git sync to GitHub or GitLab on every plan, including Free. Three things need Pro or above, starting at $25 a month: “Download code,” a custom domain, and removing the “Edit with Lovable” badge. |
| v0 | v0’s own pricing page lists GitHub sync on both Free and the $30 a user a month Plus tier. That is the more direct path. The code lands in your repository from the start, not behind a download button. |
| Bolt | Bolt’s support documentation confirms a project can connect to GitHub. The $0 a month Free plan keeps Bolt’s own branding on the finished site. The $25 a month Pro plan removes that branding and adds a custom domain. |
The practical rule: before quoting a client on a build that starts in one of these tools, open the plan you would actually pay for and read what it withholds. A prototype with the vendor’s badge still on it is not a deliverable.
Pricing at agency scale, not solo scale
Each vendor’s team tier costs meaningfully more than the one a solo builder would pick:
- Lovable scales by credits, not seats. Pro starts at $25 a month for 100 credits and rises with usage. Lovable’s own pricing page says Business adds team features on top of Pro. It starts at $50 a month for the same 100 credits.
- v0 charges $30 a user a month on Plus. Business runs $100 a user a month and adds training-data opt-out by default. That is more than three times the price for the same monthly credit allowance. The extra cost buys data handling, not more capability.
- Bolt lists Pro at $25 a month for one person. Teams runs $30 a month per member and adds centralized billing plus organization-level access management.
The decision, in one table
Use the tool when the job matches the left column. Reach for a developer, or an editor like Cursor once the code is in a repository, when it matches the right:
| Use the AI builder when | Hand it to a developer when |
|---|---|
| The client has not signed yet, and you need a visual to close the deal. | The client has signed, and the site is what they will run their business on. |
| Every form on the page is generic: name, email, message. | Any form has to reach a specific CRM, calendar, or payment account. |
| Nobody has to maintain this past the demo. | Someone is accountable for it staying up in a year. |
Once the generated repository is exported, an AI-assisted editor like Cursor (listed at $20 a month for one person on its own pricing page) is where the real build usually continues. It reads the whole repository the builder produced, which is the opposite constraint to the chat window that generated it in the first place.
What to tell the client, honestly
The tool changes nothing about running the engagement itself:
- Say which tool built the first draft. A client who later opens the codebase and recognizes the pattern trusts the agency that told them upfront more than the one that stayed quiet.
- Keep the exported repo, not just the live URL. Pull the code into your own GitHub the same day it is generated. A prototype that only exists on the vendor’s platform is one plan change away from disappearing.
- Name the maintenance owner before launch, not after. Someone has to watch the dependency, the hosting bill, and the platform’s own terms once the site is live. Decide who, in writing, before it goes live.
None of this makes Lovable, v0, or Bolt the wrong choice for a real agency. It makes them the first thirty minutes of a build, not the whole build. Know which thirty minutes they are buying before quoting a client on the rest.