Amp puts AI agents to work beyond code with OpenAI models

In an OpenAI customer story, Amp CEO Quinn Slack says OpenAI models account for most tokens used in Amp's coding agent. Amp launched Sign in with ChatGPT so users can tap more tokens and models including GPT-6 Astra without watching every increment of usage — some Amp users consume 10 billion tokens a week.

Amp uses GPT-6 Astra for documentation and design work that matches its house style, GPT-Live-1 to power Calling Puck for realtime voice control of agents including from a watch, and is testing Ultrafast mode with GPT-6 Astra.

The product insight here is honest: people who count every token never find out what an agent can do. Sign in with ChatGPT is a distribution play for OpenAI and a leash-lengthening play for Amp. Both sides win if heavy users stop rationing.

We're encouraged that Amp is pushing agents past code into planning, writing and design. That's where a lot of the unfinished work of building a product lives. The claim that Astra matches Amp's picky house style is the kind of soft quality bar that matters more than a coding benchmark for a 20-person team.

The risk is familiar: longer leashes mean more autonomous actions on a codebase or a marketing site, and mistakes scale with ambition. Voice control from a watch is convenient and also an authentication surface teams should treat carefully.

I'm watching whether Ultrafast changes how people supervise agents — fewer parallel tasks, more self-checking — or just makes the same habits go faster.