Pay-Per-Merge: The Coding Agent That Only Charges When Your Code Gets Merged

The Pricing Problem with Coding Agents

Most AI coding assistants today operate on a usage-based or subscription model — you pay for compute and API calls regardless of whether the generated code ever makes it into production. The problem is obvious: developers end up paying for a long trail of "attempts" that never land, while the portions that actually create value are nearly impossible to measure precisely.

The original source makes a pointed observation: "Coding agents usually can't price on outcomes. Ours can." This cuts right to a persistent industry pain point and hints at a new business logic beginning to take shape.

Agentic Batch Changes: Pay Only for Merged Changesets

According to the source material, Agentic Batch Changes has launched with a core selling point built around its pricing model: you only pay for changesets that actually get merged.

This means that when an AI agent processes code modifications in bulk, any changes that are rejected, fail review, or are ultimately abandoned incur no cost to the user. Only the changes that land in the codebase and generate real value are billed. It's a direct attempt to tie pricing to actual output.

What Is "Outcome-Based Pricing"?

Outcome-based pricing isn't new in traditional software and services, but it's been extremely difficult to implement in AI tools — especially coding agents. The challenge is that AI-generated output varies in quality, and verifying "whether the result worked" is itself a nontrivial problem. Code merging, however, provides a clear, quantifiable, and unforgeable success signal: the code either made it into the main branch or it didn't.

Using "whether a changeset gets merged" as the billing anchor cleverly sidesteps subjective judgments about AI output quality, and instead uses a milestone already embedded in the engineering workflow as the basis for payment.

A changeset is a fundamental concept in version control systems — it refers to an atomic set of code modifications, meaning a logically complete change that may span additions, deletions, or edits across multiple files. In modern collaborative engineering workflows, changesets are typically submitted as Pull Requests (PRs) or Merge Requests (MRs), and must pass code review and automated testing (CI/CD pipelines) before being merged into the main branch. In the context of Agentic Batch Changes, the typical scenario is: an AI agent generates a batch of modification proposals targeting multiple codebases or multiple areas of technical debt in one pass, with each proposal corresponding to one changeset. This kind of bulk operation is especially common in large-scale refactoring, dependency upgrades, and security vulnerability fixes — tasks with historically high manual labor costs, and precisely the kind of work AI agents are built to handle.

Implications and Limitations of This Model

From a business model perspective, pay-per-merge aligns the vendor's incentives with the user's actual outcomes. The vendor is motivated to improve the agent's code quality and merge success rate, while users face lower trial-and-error costs and a lower barrier to adoption.

That said, there are boundaries worth watching. For instance, a merge decision is influenced by external factors like team review standards and project conventions — it doesn't purely reflect the AI's technical capability. Additionally, how a single "changeset" is scoped and defined will directly affect the fairness of billing.

Given the limited public information available at this stage, these details still await further clarification from the product team and real-world validation.

A useful analogy comes from outcome-based contracts already practiced in software outsourcing and SaaS. Some RPA (Robotic Process Automation) vendors charge per successfully processed business transaction; some legal tech companies take a cut only on won cases. These precedents show that tying pricing to verifiable results effectively reduces procurement risk for buyers — but it also demands stronger risk-bearing capacity and attribution mechanisms from vendors. What makes AI coding agents uniquely complex is that the merge decision is made by a third party (the development team), meaning the vendor doesn't fully control the final outcome. This makes the risk-sharing design more intricate than traditional outcome-based contracts.

What This Means for Developers

For teams evaluating AI coding tools, this pricing innovation is worth paying attention to. It reflects a broader evolution of coding agents — from "pay-per-use tools" to "value-delivery partners." When choosing a tool, developers might add "does the pricing reflect real output?" to their evaluation criteria. That question often reveals more about a vendor's confidence in their own product than any feature comparison chart.

Whether outcome-based pricing becomes the dominant paradigm for coding agents depends on both technical maturity and market acceptance. But at the very least, it's an intriguing signal — one where a vendor puts its confidence right on the price tag.