AutoGPT

Overview AutoGPT is the project that gave the industry its first honest look at what "autonomous agent" was going to feel like. It shipped in March 2023 as a Python script that wired GPT 4 to a scratchpad, a file system, and a browser, then let the model loop on its own until it decided the task was done. Within weeks it had more than a hundred thousand GitHub stars and a running argument about whether the future had arrived or whether we had just built a very expensive way to make an LLM google itself forever. Both were partly true. AutoGPT Classic — the original script — was more demo than product. It hallucinated tool calls, burned through OpenAI credits like a space heater, and ended most sessions in a loop that only a keyboard interrupt could break. But the pattern it introduced — plan, act, observe, re plan, until goal — is the pattern every serious agent framework in 2026 still follows, and the AutoGPT project itself has since rebuilt around a real platform rather than a viral notebook. In 2026, "AutoGPT" mostly means the AutoGPT Platform at agpt.co — an open source, self hostable, low code agent building environment with a visual block editor, a marketplace of prebuilt agents, and a scheduler that lets you run agents on a cron. The framework classic still exists in the repo history and a few community forks, but the maintained product is the platform. If you're evaluating AutoGPT today, that's…