Akka published an account of using a spec-driven AI delivery harness to rebuild 65 popular open source projects, including Uptime Kuma, Changedetection.io, Redash, Glance and ActivityWatch, on its own platform in a reported 99.3 hours across an initial tranche at a cost of about 9.41 billion tokens. The company said it selected projects across popular AI frameworks, Go and TypeScript networking libraries and complex software-as-a-service applications, then ran a loop of setup, discovery, porting, benchmarking and improvement in which Claude with Akka Specify handled implementation, testing and review while a common benchmark runner compared behavior, code size and latency. Akka said it licensed the ports under Apache licensing except where GUI interfaces were copied verbatim and kept under the original terms with attribution, and it reported mixed results, with some rebuilds much smaller and faster and others, such as Netflix Metaflow, roughly 100 times slower, while its own harness notes that the workloads compared for the largest claimed speedup differed.
Akka says it used a spec-driven AI harness to port 65 open source projects
Akka published an account of using a spec-driven AI delivery harness to rebuild 65 popular open source projects, including Uptime Kuma, Changedetection.io, Redash, Glance and ActivityWatch, on its own platform in a reported 99.3 hours across an initial tranche at a cost of about 9.41 billion tokens. The company said it selected projects across popular AI frameworks, Go and TypeScript networking libraries and complex software-as-a-service applications, then ran a loop of setup, discovery, porting, benchmarking and improvement in which Claude with Akka Specify handled implementation, testing and review while a common benchmark runner compared behavior, code size and latency. Akka said it licensed the ports under Apache licensing except where GUI interfaces were copied verbatim and kept under the original terms with attribution, and it reported mixed results, with some rebuilds much smaller and faster and others, such as Netflix Metaflow, roughly 100 times slower, while its own harness notes that the workloads compared for the largest claimed speedup differed.
Source: Akka