Amp Expands AI Agents Beyond Coding
- •Amp uses OpenAI models to expand its agent’s work beyond coding into planning, writing, design and marketing
- •Sign in with ChatGPT gives Amp users more tokens and access to GPT-6 Astra; some consume 10 billion tokens weekly
- •Puck can coordinate agents, while Calling Puck uses GPT-Live-1 for real-time voice and Ultrafast mode speeds workflows
Amp, a North American startup, is using OpenAI models in its AI agent for work beyond coding, including deciding what to build, writing, designing pages and marketing. CEO Quinn Slack says OpenAI models account for the majority of tokens used in Amp, which recommends models based on what works best for its users. Amp’s 20-person team also uses an agent to improve the experience for new users and help people discover the product through search and other channels.
Amp launched Sign in with ChatGPT to give users access to more tokens and OpenAI models, including GPT-6 Astra. Slack says some Amp users consume 10 billion tokens a week, and that the integration can make users less concerned about tracking every token and more willing to assign agents larger tasks. He says users also use agents to plan, decide what to build and continue working after a product ships.
Amp says GPT-6 Astra helps its team produce documentation and design pages that match its writing and design standards. Slack also used plans for his house to create a 3D model in Blender, which he and his children turned into games. He said that use was not possible just a couple months ago.
Amp recently announced Puck, a conversational assistant built into Amp that can launch and manage other agents, coordinate several agents, investigate bugs and find past conversations. Calling Puck, powered by GPT-Live-1, lets users talk to Puck in real time, including from a watch. Amp has also been testing Ultrafast mode with GPT-6 Astra; Slack says faster responses may change how users juggle agent tasks or give agents more time to check their work. Amp says its next step is to bring more work before and after coding into its agent.