Agentic Layer Check
Lovable prices the one thing most AI vendors hide: how much work the model actually did. A credit charge rises and falls with the complexity of the request, a half finished build is billed at half, and a request stopped before anything changed costs nothing. Then the goal command hands the agent the decision about how much work to do. Here's how that reads against the nine layers.
What they sell
An AI app builder: a chat interface that writes, edits and deploys a working web application, with hosting, a cloud backend and AI features bundled in behind the same balance.
Who they sell to
Founders, product people and non engineers building and shipping apps themselves, self serve from Free through Pro and Business, plus departments and larger organisations on Business or custom Enterprise.
How they price
Credits, charged per message and varying with the work done. Free (daily allowance only), Pro (from $25/mo) and Business (from $50/mo) both scaling with credit volume into the thousands, plus custom Enterprise. Unlimited workspace members on every plan, with top ups available on Pro and Business.
Two choices here are worth crediting properly, because very few vendors in this category make either one.
Lovable's own documentation states the rule plainly: "Cost depends on the complexity of the request and the work completed." Its own pricing page prices that out in public, from around half a credit to restyle a button up to roughly two credits to build a landing page with images, and the cost of every message is visible in the message history. Most AI native tools do the opposite, flattening wildly different amounts of model work into one opaque credit rate, which leaves the customer unable to answer the only question that matters: what does one credit actually buy me. Lovable answers it per message, before and after the fact.
Per Lovable's documentation, "Stopped Build mode requests are charged based on the work completed so far. A request you stop before Lovable has made any change to your project is not charged." That is a genuine entitlement decision, not a billing detail. It puts the risk of a wasted run on the vendor rather than the customer, in a category where a model that goes off course is a normal Tuesday. Pair it with unlimited workspace members on every plan and credit top ups on both paid tiers, and the packaging stays out of the way of the thing the customer is trying to do.
A meter this honest about work done creates a second question: who decides how much work gets done.
Lovable's documentation is candid about the goal command: a message set as a goal "is also billed on the work Lovable does, but costs noticeably more: the same work uses more credits, and Lovable keeps working until the goal is achieved." Every word of that is defensible, and together they describe a commitment made before the price is knowable. The customer approves an outcome; the agent decides how many credits reaching it takes. That inverts the usual direction of a usage model, where the customer's own actions set the bill, and it means revenue per customer is now partly a function of how hard their goals happen to be, which is not a variable Lovable or the customer sets in advance. The opportunity is a ceiling, an estimate, or a mid run checkpoint: something that lets a customer commit to an outcome and a maximum at the same time.
The number on screen is a single credit balance. Behind it, per Lovable's own documentation, monthly plan credits roll over but still expire two months after they were issued, while usage specific grants do not roll over at all, and daily build credits, the monthly Cloud grant and the monthly AI grant each run on their own clock. A customer watching one number cannot tell which part of it is about to disappear, and finance on the vendor side has to model four different expiry behaviours to forecast what that balance is worth. Showing the customer which credits expire when, inside the same view, would turn an accounting structure into a reason to keep using the product.
The goal run doesn't stay a Layer 8 question. It lands on Pricing & Packaging (Layer 3), because a plan sold on a credit volume is being asked to carry a workload whose size the buyer cannot estimate at purchase, which is exactly when a customer over buys once and then under uses for a year, or under buys and feels rationed. It lands harder on Adoption & Customer Success (Layer 7): the feature most likely to produce a customer's best outcome is also the one most likely to produce a surprising bill, so the success motion has to teach customers when to spend more rather than wait for them to discover it on an invoice. That's the same pattern this series keeps finding. Lovable made the most honest metering decision in its category, and the cost of that honesty shows up two layers away, in how the plan is sized and how customers are taught to use the thing that costs the most.
Part of the Agentic Layer Check series: named companies, read against the nine layers. Want your own read? Request a teardown →
This is a public signal read: Lovable's own pricing page, its own credits and usage documentation, and its own subscription plan documentation, not a private briefing or company input. It reflects what's publicly documented as of the date above: pricing pages and credit terms change, and this piece will note it if Lovable's does. Lovable was not consulted on or paid for inclusion; see the disclosure approach this work is run under.
Sources: Lovable: pricing → · Lovable docs: credits and usage → · Lovable docs: subscription plans →