Overview
Magic is a San Francisco-based AI company building an AI software engineer that acts as a collaborative teammate for software development, capable of handling large end-to-end tasks with minimal human supervision and aiming for artificial general intelligence. Its approach combines frontier-scale pre-training, domain-specific reinforcement learning, ultra-long context, and high inference-time compute, centered on a Long-Term Memory model that can process very large codebases. The company differentiates itself by enabling end-to-end software development with a memory-equipped agent rather than just a tool, leveraging a very large context window to understand entire projects. Its goal is to automate complex software engineering tasks and advance toward AGI in the software development domain.
Work here?Claim your company
Simplify's take
What believers are saying
- Magic said in September 2026 its pretraining became over 10x more compute-efficient.
- Revelio counted 98 employees in June 2026, up 15.5% year over year.
- Magic disclosed $768 million raised by September 2026, supporting continued infrastructure spending.
What critics are saying
- Magic had no shipped product or revenue by October 2026.
- Its September 2026 claims lacked independent verification, keeping technical credibility fragile.
- If Anthropic, OpenAI, or GitHub ship reliable agents first, Magic loses relevance.
What makes Magic unique
- Magic targets an autonomous software engineer, not a copilot, from San Francisco.
- Its LTM-2-mini claimed a 100 million-token context window in August 2024.
- Magic combines frontier pretraining, domain-specific RL, long context, and inference-time compute.
Help us improve and share your feedback! Did you find this helpful?