Magic

Magic is an AI research company founded in 2022 by Eric Steinberger and Sebastian De Ro, headquartered in San Francisco, California. The company specializes in developing AI models designed to automate various software engineering tasks, including code generation, review, and debugging. Their proprietary Long-Term Memory (LTM) architecture enables models to process ultra-long context windows, handling up to 100 million tokens—approximately 10 million lines of code—in a single context window. This capability allows Magic's models to understand and work with entire codebases simultaneously, facilitating more efficient and comprehensive software development processes.

In August 2024, Magic released LTM-2, their first publicly demonstrated model, showcasing the practical application of their LTM architecture. The company has attracted significant investment, raising over $465 million from notable investors such as former Google CEO Eric Schmidt, Sequoia Capital, Alphabet's CapitalG, Atlassian, Jane Street, Nat Friedman, Daniel Gross, and Elad Gil. This funding underscores the confidence investors have in Magic's approach to advancing AI-driven software engineering solutions.

Magic's flagship product, the LTM-2-mini model, is capable of processing 100 million tokens in a single context window, enabling it to manage entire codebases effectively. This model is part of Magic's broader strategy to develop autonomous coding agents that can perform complex software development tasks with minimal human intervention. The company's focus on ultra-long context windows addresses critical bottlenecks in computational efficiency and data governance, supporting broader industrial automation.

The company has established a partnership with Google Cloud to build two AI supercomputers, Magic-G4 with NVIDIA H100s and Magic-G5 with Blackwell chips, to support their AI research and development efforts. Magic operates in a competitive market, with competitors including Codeium, Cognition, Poolside, Anysphere, and Augment. Despite the competition, Magic's unique approach to integrating ultra-long context windows into AI models positions it as a significant player in the AI-driven software development sector.