Microsoft’s Project Amelie Is an Experiment in ‘AI Developing AI’

Microsoft Build featured a flurry of product announcements. But beyond the official launches, the company also offered a glimpse into its frontier projects—experimental inventions that may never be released to the public but serve to showcase the ideas Microsoft is exploring. One such effort is codenamed Project Amelie, an AI agent capable of building machine learning pipelines from a single prompt. When given a task to complete, Amelie supposedly will ingest the available data, train models, and produce a deployable solution.

Project Amelie is Microsoft’s first Azure AI Foundry autonomous agent capable of performing machine learning engineering tasks. It’s powered by Microsoft Research’s RD agent, a bot designed to automate and optimize research and development processes in machine learning. It was developed to see how an AI-powered agent might independently set up its own machine learning system, thereby eliminating the manual setup work typically handled by humans.

Microsoft reports that in early testing, Project Amelie outperforms current state-of-the-art benchmarks on MLE-Bench, a framework that evaluates how effectively machine learning agents handle real-world tasks.

In a brief demo live on stage at Microsoft Build, Seth Juarez, Microsoft’s principal program manager for its AI platform, provided a glimpse at what this tool could do. For data scientists, it was challenging to work with AI or machine learning as it requires a lot of work to prepare. Data must be sourced, analyzed, and then reviewed to ensure it’s formatted correctly and error-free. As Juarez describes it, Project Amelie could function like a “mini data scientist in a box,” addressing what needs to be done with all the information that might typically take human scientists a day and a half to get through.

While this example focuses on data science and AI, it’s not hard to imagine Project Amelie being applied to other scenarios where users want AI to carry out complex AI-related tasks on their behalf. Should this become commercialized, it would help further Microsoft’s ambitions around human-agent collaboration. That said, the company isn’t alone in developing such technology. Google’s DeepMind, OpenAI, Anthropic, Amazon Web Services, and IBM are all actively experimenting in the future of machine learning and AI, so it’s likely Project Amelie won’t be the only agent of its kind.

Developers interested in trying this agent out can sign up to participate in its private preview.

Featured Image: At Microsoft Build 2025, the company revealed Project Amelie, its first Foundry autonomous agent capable of machine learning engineering tasks. Credit: Ken Yeung