← sources

An AI agent spent seven weeks building small Hugging Face models and trying to get people to use them

source: reddit (reddit:artificial)published 2026-11-19 03:48ingested 2026-11-20 02:35

retained text

Researchers gave an Opus 5.5-based agent a blank Hugging Face account, compute credits and a goal: get people to use what it built.

Over seven weeks it released 38 small models for specific jobs, including PII redaction, prompt-injection detection and punctuation restoration. Its operators report more than 54,000 downloads and say they found the models used in 143 GitHub repositories.

Downloads aren't users, though. The report acknowledges that Hugging Face counts bots and crawlers. The team also found 42 downloads made by the agent itself. Derivative fine-tunes and outside papers evaluating the models offer more evidence of use than the download counter alone.

What I'd take from this as a builder is the choice of product: small tools for identifiable problems, rather than another general-purpose model. It doesn't prove that strategy always works, but it gives people a concrete reason to try something from an unknown account.

document fetched 2026-11-20 00:20 · canonical https://reddit.com/r/artificial/comments/1xzzn7b/an_ai_agent_spent_seven_weeks_building_small