Costs · October 19, 2026

What Vibe Coding Really Costs a Service Business in 2026

Subscriptions, hosting, credit burn and your own hours. Three realistic monthly budgets for building an AI tool with Lovable, Bolt, Replit or Claude Code.

The pitch for vibe coding is that software got cheap. A $25 subscription, a weekend, and you have an app. For a prototype, that is close to true.

For a tool your business depends on, or one your clients pay for, the subscription is the smallest number on the bill. This post walks through what the tools cost as of October 2026, where the surprise charges come from, and three realistic budgets for a service business. It ends with the cost most owners leave out entirely: their own hours.

The sticker prices

Here is what the main tools charge on their entry and mid tiers, as of October 2026. Prices change often, so check each pricing page before you commit.

Two lines in that table catch business owners out.

First, Vercel's free Hobby plan is for non-commercial use only. If the app serves your business or your clients, you are on Pro.

Second, Supabase's free tier pauses your project after a week without activity. That is fine for an experiment. It is not fine for a client portal that someone opens on a Monday after a quiet week.

Lovable and Bolt bundle hosting, so you can publish without a separate Vercel bill. A custom domain on either needs a paid plan. Claude Code, by contrast, does not host anything. It writes the code, and you pay separately for somewhere to run it. Our post on where a vibe-coded app actually lives covers that in detail.

Where the real money goes

None of those subscriptions will hurt a business that bills five or six figures a month. The costs that hurt are the variable ones, and they grow at exactly the moment you are least able to watch them.

Credits burn fastest when things break

Most builders charge per message, per credit or per token. The first build of an app is relatively cheap. The edits afterward are not, because every change means the AI has to read and reason about an app that keeps getting bigger.

The worst case is the fix loop. Something breaks, you click "try to fix," the fix breaks something else, and you click again. Reviews of Lovable describe people spending dozens of credits undoing bugs the AI itself introduced, and anecdotes run from a couple of hundred dollars to a few thousand spent trying to get back to a working version. Treat those numbers as anecdotes, but the pattern is consistent across tools.

The standard advice holds: when a fix goes sideways, roll back to the last working version instead of asking the AI to fix its fix. Use a plan or chat mode to diagnose before you let it change anything.

Effort-based pricing is hard to forecast

In September 2025, Replit launched Agent 3 on effort-based pricing, where each task is billed by how much work the agent decides to do. The Register reported users seeing their costs jump: one spent around $1,000 in a week after previously paying $180 to $200 a month, another lost $70 in a single night. Replit acknowledged that effort-based pricing "can end up being more expensive over the lifetime of a project."

The lesson is not that Replit is bad. It is that any tool billing by agent effort turns your budget into a forecast you cannot make. Edits to an existing app were the expensive part for those users, and an app in production is mostly edits.

Subscriptions have ceilings too

Flat-rate plans protect you from runaway bills, but they cap how much you can do. Anthropic added weekly usage limits to Claude plans from August 28, 2025, on top of the existing five-hour windows. Most subscribers never hit them. A non-technical owner leaning on Claude Code to debug a growing app can, and then the choice is to wait or upgrade to Max at $100 or $200 a month.

The AI inside your app costs money every time a customer uses it

This is the line item that grows with success. If your tool reads documents, drafts replies or answers client questions, each of those calls goes to a model provider and is billed by usage. Ten internal users barely register. Two thousand clients using it every week is a real monthly cost, and it scales with exactly the thing you are trying to grow.

Simon Willison's advice is the single most useful sentence in this post: be very careful vibe coding against usage-billed APIs, and set a billing limit. He cites people who ran up thousands of dollars because nothing stopped them. Every provider you connect, whether for AI, maps, email or SMS, should have a hard spending cap before your app goes live.

A leaked key is a bill someone else runs up

In March 2025, a founder who had built a SaaS entirely with Cursor shut it down within a week after attackers found API keys exposed in the front-end code and maxed them out. Security is a cost question as much as a safety one. Our security checklist covers how to avoid it.

Three realistic monthly budgets

These assume you are a service business, not a hobbyist, and that you set billing caps on everything.

1. The weekend prototype: $0 to $50

What it is: A clickable version of your idea to show your partners, your team or three friendly clients. Fake data, no real logins, nothing sensitive.

What you pay for:

• One builder subscription, such as Lovable Pro or Bolt Pro at $25, or Claude Pro at $20 if you are comfortable with a terminal.
• The free tiers of everything else.

Watch for: Nothing much. This is where vibe coding is genuinely excellent. Spend the money, cancel after, and you have learned more about your idea than a month of meetings would teach you.

2. The internal tool: roughly $75 to $250 a month

What it is: Something your own staff use every day. A proposal generator, a scheduling helper, a dashboard that pulls from your job management system.

What you pay for:

• A builder or Claude subscription for ongoing changes: $20 to $50.
• Supabase Pro so the database does not pause: from $25.
• Vercel Pro if you host outside the builder: $20 per user who needs access.
• AI usage inside the app: small at internal volumes, but capped.
• Occasional top-up credits when a change gets complicated.

Watch for: Real company data is now in the system. Staff details, client names, maybe financials. The tool bill is still modest. The cost of a data leak is not.

3. The client-facing product: $250 to $1,000+ a month, before your time

What it is: A portal, assistant or self-serve tool your clients log into, and ideally pay for.

What you pay for:

• Higher builder or Claude tiers, because you will be making changes constantly: $50 to $200.
• Database and hosting on paid plans, likely more than one environment so testing does not touch real client data: $50 to $150.
• AI usage that scales with client activity: anywhere from tens to many hundreds of dollars.
• Email, SMS, payments and monitoring services, each with its own small bill.
• A professional security review before launch. This is a one-off, and it is worth getting a quote before you decide you can skip it.

Watch for: Every cost in this tier rises with usage. That is fine if the product earns more than it costs, which is the whole point. It is a problem if you priced it as a free add-on to your service.

The cost nobody puts in the spreadsheet: your hours

Jason Lemkin, who built five production apps at SaaStr without a developer, wrote up what it actually took. His guidance: budget about a month for a real B2B app, and expect around 60% of that time to go on testing and fixing.

Now put a number on it. Say you are a partner who bills $250 an hour, and the app takes 120 hours across six weeks of evenings and Fridays. That is $30,000 of your time, against a software bill of a few hundred dollars.

It does not end at launch. Somebody has to answer when a client says the login stopped working. Somebody has to update the app when a provider changes its API. If that somebody is you, those hours come out of the work that pays for everything else.

None of this means you should not build it yourself. For prototypes and internal tools, the trade is often excellent. It means the honest comparison is not "$25 a month versus hiring someone." It is your billable hours, plus the variable costs, plus the risk, versus the alternatives. We go through those in build it yourself or hire.

How we think about the bill

We tell owners to vibe code their prototype. It is the cheapest way to find out whether an idea holds up, and you will understand your own product better for having tried.

Where we come in is the step after, when the tool is meant to face clients and earn money. You get a working prototype from us before any money changes hands. If it lands, there is one fixed build fee, agreed after you have clicked through it. After that we are paid as a share of what the product earns, and we stay on as its product lead, so the ongoing hours that would otherwise come out of your week are ours.

That structure exists for the reason this post does. The subscription is never the real cost of software. The real cost is what it takes to keep it earning, and we would rather be paid out of that than out of your calendar.