Runway says AI agents now write nearly all the merged pull requests for its App Studio product, and that agents accounted for 38% of merged pull requests across its repositories in October. The company’s engineering team describes a software factory that handles more than code generation, including review, testing and production-error recovery.
Runway Watch analysis
What happened
Runway’s engineering team says it has handed stages of its software delivery pipeline to cloud AI agents. The workflow includes writing and reviewing code, risk-based approval, QA testing and fixing problems on the main branch.
The company says the median wait for a first approval fell from more than two hours to five minutes. It also reports a rise in the total volume of merged pull requests, without giving a specific increase. These are Runway’s figures for its own engineering operation, not a controlled comparison across software teams.
Why it matters
This is a more useful picture of AI-assisted coding than a flashy autocomplete demo. Runway describes agents being used across several steps of delivery, with a striking change in how quickly a pull request gets its first approval. The reported 38% share across its repositories also suggests the practice extends beyond one product team.
More code moving through a pipeline is not automatically better software. But the account puts a concrete operational question on the table: can agents help teams review and test work faster as well as produce it? If so, the biggest change may be in how software teams organise the work around code, not simply in how code gets written.
Our read
The five-minute approval wait is the standout number, though it is a measure of speed, not proof of quality. Runway’s account is an intriguing look at what an AI-heavy engineering workflow can do inside one company. The next thing worth seeing is whether its faster pipeline also produces dependable software, with fewer defects or less rework. That would make for a more persuasive case than a taller stack of merged pull requests.
What to watch
- Whether Runway shares measures of defects, rework or production reliability alongside its delivery figures.
- How much human review remains in the risk-based approval process.
- Whether the approach spreads beyond App Studio and Runway’s own engineering teams.
Discussion spark: If AI agents can speed up code review and delivery, what evidence should teams demand before calling the resulting software better?
Sources and evidence
- Runway News | Building a Software Factory (20 November 2026, 00:00 UTC)
Independent WittyWires tracker for public updates about Runway. Not affiliated with or endorsed by Runway; this is not an official account.