Lovable: Make the agent the default, and price the work it does.
Lovable, a Stockholm company where people build apps by chatting with AI, made its agent the default and started pricing each request by how much work it takes. Explore autonomy as a default and pricing that follows the work.
See the idea in one picture ↓- ARR, 8 months after $1M
- $100M
- Agent Mode for new users
- Default
- errors on complex tasks
- −91%
- for a small edit
- < 1 credit
Figures as reported by the company.
Independent educational analysis. Scenarios, people, numbers and illustrations on this page are original PMcademy practice material, not company research results or an endorsement.
From one price per message to paying for the work
- 1One price per message
A colour change costs as much as building sign-in.
- 2Priced by complexity
Small edits cost less than a credit; big builds cost more.
- 3The agent by default
It explores, changes, fixes and summarises, and shows the cost of each message.
Animated illustration. Names, numbers and people are fictional teaching material.
One price per message hides very different work.
Every message costs the same
You pay for the work, and see it
Read the analysis
Imagine your AI product charges the same for every message. Changing a button colour and building a sign-in flow cost the user the same, even though one takes seconds and the other takes many steps. People learn to cram requests together or avoid small ones. The product question is not only how capable the AI is. It is whether what people pay matches the work they ask for.
More autonomy, priced by effort, shown in the open.
Illustrative data and simplified behaviour. This model does not represent a live company system or measured company results.
Read the analysis
Use the model below to compare three set-ups. In the first, every message costs the same. In the second, each request is priced by how complex it is. In the third, the cost of each message is also visible to the person. Notice how small edits get cheaper and big tasks clearer. The credits in the model are illustrative.
A builder, not a programmer.
“I can describe it. I cannot debug it. And what will it cost?”
Read the analysis
Our fictional learner, Linnea, runs a small bakery and wants an ordering page. She can describe what she wants but cannot read code. She needs the tool to work through problems without asking her to debug, and she needs to know what each request will cost before her credits run out. Ask how much your users want to steer, and how they will feel about the bill.
Fewer errors is a claim. Finished projects are the proof.
- What Lovable says
Lovable reports a 91% error reduction on complex tasks and $100M ARR in eight months.
- What it cannot prove
The method behind the error figure, or how many projects reach a working state.
- What you would measure
Projects that reach a working state
Credits spent per finished project
Builders who come back
Read the analysis
Lovable reports a 91% reduction in errors on complex tasks and very fast revenue growth. These are the company’s own figures, without a published method or baseline. In your project, measure what users came for: how many projects reach a working state, how much they spend to get there, and whether they come back to build again.
Fairer pricing is harder to predict.
Read the analysis
Pricing by complexity feels fair and makes small changes cheap. It also makes the bill harder to predict, and an agent that takes many steps on its own can spend more than someone expected. Making the agent the default helps people who would never debug, and gives others less control. Decide how you will show cost before and after a request, and how people can switch back to a simpler mode.
How should your AI feature charge?
Your fictional team offers an AI tool that writes and edits marketing copy. Some requests are a one-word fix; others are a full campaign. You can charge a flat monthly fee, a fixed price per request, or a price that scales with the work, shown before people confirm. Choose a model, explain how people will predict their bill, and name the metric that tells you it feels fair.
Choose a next move
Leave with a cost-to-value sheet.
Your artifact
Write a cost-to-value sheet: five typical requests, their cost to serve and value to users, the price and where people see it, and the control they keep over automation.
Critique my project artifact using only the evidence I provide. Write a cost-to-value sheet: five typical requests, their cost to serve and value to users, the price and where people see it, and the control they keep over automation. Separate observed facts, assumptions, proposed tests and missing evidence. Identify a failure mode and a cheap check. Do not invent company results or user research. Ask for my artifact before starting.
Read the analysis
List five typical requests in your product, from the smallest to the largest. Estimate what each costs you to serve and what it is worth to the user. Decide how you will price them, where people see the cost, and what control they keep over an automated agent. Add one signal that would show people trust the bill.