Amp broadens its AI agent workflow with OpenAI models
Amp is expanding its AI agent from software development into planning, design, documentation, marketing and post-launch work, using OpenAI products including GPT-6 Astra, GPT-Live-1, Ultrafast mode and Sign in with ChatGPT. The North American startup says OpenAI models account for the majority of tokens used in Amp, while some individual users consume 10 billion tokens each week.
Quinn Slack, Amp’s CEO and member of technical staff, says model improvements let the company extend the agent’s role beyond writing code. Amp recommends models based on what works best for users, and Slack cited OpenAI’s transparency when safeguards are applied as a factor that makes the models easier to deploy in the product.
Token access changes how much work users delegate
Amp introduced Sign in with ChatGPT to give users access to more tokens and OpenAI models, including GPT-6 Astra. Slack said high token consumption can make builders cautious about assigning larger jobs to an agent, whereas broader access can encourage them to let the agent work through longer tasks.
That matters because Amp users are asking agents to plan work, decide what to build and continue after software is shipped. The product direction aligns with agents that carry tasks through to outcomes as organizations assess agents that can carry tasks through to an outcome rather than merely return an answer.
From written output to live coordination
Amp’s own 20-person team uses its agent outside engineering, including for marketing, improving the new-user experience and helping potential users discover the product through search and other channels. Slack said GPT-6 Astra has helped the team produce documentation and design pages that match Amp’s particular writing and design requirements.
The company has also announced Puck, a conversational assistant inside Amp that can start agents to investigate a bug, coordinate multiple agents or retrieve an earlier technical conversation. Calling Puck, powered by GPT-Live-1, enables real-time voice interaction with the assistant, including from a watch.
Speed alters agent workflow choices
Amp is testing Ultrafast mode with GPT-6 Astra. Slack said faster responses may reduce the need for users to switch among several agent tasks while waiting. It could also leave agents more time to check their own work before returning a result.
For businesses evaluating agent tooling, Amp’s approach points to a practical test: assess agents across planning, production, coordination and review, while ensuring token access, quality standards and human oversight fit the work being delegated.