Amp is a coding agent for terminal and editor workflows, spun out from Sourcegraph to operate as an independent research lab. The product enables developers to delegate complex coding tasks to an AI agent while maintaining human control over direction and review - a constraint that shapes both the interface design and the interaction model.
The agent operates across multiple modes (smart, rush, deep) and supports multi-model inference, allowing users to route tasks to different models based on latency and capability trade-offs. A threads system provides persistent, shareable execution records within teams, reducing the operational friction of knowledge transfer across sessions and enabling asynchronous collaboration on coding problems.
Amp is profitable and accessible from both terminal and editor environments, positioning it to integrate into existing developer workflows rather than requiring context switching. The team prioritizes shipping and experimentation in the product itself - infrastructure choices and interaction patterns grounded in real usage rather than speculative architectural claims.