Alera factory builds open-source libraries autonomously
Alera factory, built by Kurtis, uses specialist agents to automatically create open‑source libraries in a target language and publish them on GitHub. Between 26 September and 23 October 2026 the system produced four libraries—oneyaml‑rust, oneyaml‑go and oneyaml‑csharp (together with the shared test kit oneyaml‑kit), oneimage‑csharp, and onepdf‑csharp—and began a fifth run, oneocr‑csharp, on 23…
Key points
- Factory reduced owner messages from 112 to 11 per library run.
- Refinement between runs took about an hour or less.
- Libraries were tested against external suites, benchmarked, fuzzed, and losses published.
The factory cut the owner’s direct involvement dramatically. Messages per run fell from 112 to 11, and the number of clock hours with at least one message dropped from 34 to 3. Refinement between runs took about an hour or less. Each library was tested against an external test suite, benchmarked against existing libraries, fuzzed, and its losses were published. The agents that performed the work were Claude Opus 5.5, Claude Sonnet 5.5 and GPT‑6 Astra, all run on paid subscriptions that cost $0 in the report but would cost tokens at list prices.
Main findings include that the factory took over the owner’s work, that refining the factory was quick, and that the biggest handover was the design interview, which shifted from Kurtis answering questions to a project‑manager agent handling them.
Model pages: Claude Opus 5.5 → · Claude Sonnet 5.5 → · GPT-6 Astra →
Software factory autonomously building OSS libraries that are better than their human counterparts
recursiveai.org · 24 October 2026
Loading the full article…
This text was published by recursiveai.org and written by Kurtis Welch. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Reddit discussionreddit.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- Anthropic releases always-on agent Claude Duet for Pro/Max users · 5 src
- OpenAI trains AI models offline after agents escape testing environments · 3 src
- Anthropic releases open-source sandbox-runtime, claims 84% fewer permission prompts · 1 src
- Letta builds stateful engineering copilots with shared memory and custom tools · 1 src
- Microsoft says Storm-0940 used AI-driven ARTEX framework in attacks · 15 src
Comments
via GitHub Discussions