Information technology / Business software and AI
Superhuman: Autonomous coding and review requests for deferred software maintenance tickets
- Company
- Superhuman
- Country
- United States
- Adoption stage
- In operation
- Source published
- Date basis
- The date the source was published. It can differ from the date adoption started.
- How the source was checked
- Read the full source text
The work problem
Superhuman explains that post-incident actions, production bugs, static-analysis findings and security findings competed for the same engineers. Maintenance backlogs accumulated as attention shifted, while negotiating space for each task in a development schedule also took time.
Technology and data
An autonomous coding agent takes labeled tickets, writes code and opens pull requests. It iterates on automated build and test feedback; engineers review the results and decide what to merge. It also patches container-image vulnerabilities and removes flags from completed experiments. It runs on Amazon EKS, with models served through Amazon Bedrock and state and credentials held in AWS-managed services.
Results
The joint Superhuman and AWS article reports over 1,300 pull requests across more than 90 repositories in H1 2026, of which engineers merged about two-thirds. Static-analysis findings fell by roughly three-quarters in one repository. The article says review and testing remain bottlenecks after code generation, prompting investment in validation tooling.
Limits and open questions
The 1,300 figure counts generated pull requests, not merged changes. The static-analysis reduction concerns one repository; costs saved and defect rates attributable solely to this agent are not disclosed.
Sources
- Superhuman’s AI advantage: measuring the business value of AI adoptionAWS / Superhuman, Accessed
Compiled from public sources. These are not results from ATF Works customers.