BabyAGI was published in early April 2023 by Yohei Nakajima, an investor who described himself as someone who had never held a job as a developer. It grew out of a Twitter thread about a “task-driven autonomous agent” and landed as a strikingly short Python script - on the order of a hundred lines. That brevity was the point: it made the core loop of an LLM agent legible to anyone who could read code.
The original BabyAGI ran a simple cycle. It pulled the highest-priority task from a list, used a language model to complete it, used the model again to create new tasks based on the result and the overall objective, and reprioritized the task list, storing results in a vector database so context could be retrieved later. Execution, task-creation, and prioritization were each just a prompt. Released days after AutoGPT, it became the other reference implementation everyone pointed to when explaining how autonomous agents worked.
BabyAGI’s influence came from its minimalism. It was easy to read, fork, and extend, so it seeded countless tutorials, variants, and production experiments. Nakajima later rebuilt the project around a framework for self-building agents and archived the original in 2024, but the early script remains a canonical teaching example of the task-loop that defines an LLM agent.