Amp’s AI Agent Platform
Amp is developing an AI agent platform that leverages OpenAI models to assist developers in transforming ideas into functional software. The company strategically selects models based on user needs, with OpenAI models accounting for the majority of tokens utilized on the platform.
Reducing Token Friction with Sign in with ChatGPT
Quinn Slack, Amp’s CEO, noted that some users consume as many as 10 billion tokens per week. To mitigate cost and usage constraints, Amp has integrated Sign in with ChatGPT. This integration provides users with access to additional tokens and OpenAI models, including GPT-6 Astra. Slack emphasized that “Sign in with ChatGPT means that people can have that feeling where they’re not counting every single token,” allowing users to assign larger tasks to agents without the constant worry of token consumption.
Writing, Designing, and Real-Time Interaction
Amp’s own 20-person team utilizes agents for a variety of tasks, from product planning to marketing. GPT-6 Astra plays a crucial role in producing documentation and designing pages that align with the company’s specific style requirements. Slack shared a personal example of using GPT-6 Astra at home to convert house plans into a 3D model in Blender, which his children then used to create simple games.
Puck: Managing Other Agents
Amp has introduced Puck, a conversational AI assistant that facilitates the launching and management of other agents. Users can instruct Puck to initiate an agent for bug investigation, coordinate multiple agents, or retrieve past technical conversations. This functionality enhances the overall efficiency of managing tasks within the platform.
Accelerating Workflows with Ultrafast Mode
Amp has also been testing an Ultrafast mode with GPT-6 Astra. Slack mentioned that the increased responsiveness changes how users interact with agents, potentially allowing them more time to verify their work before returning results. He stated, “People have not seen a model that’s this good, that’s this fast.” Current efforts are focused on expanding agent responsibilities both before and after coding, enabling builders to advance their ideas without needing to manage every step themselves.
Original source: OpenAI News