News
Amp uses OpenAI models so agents work past the editor
OpenAI’s Oct. 14 case study says Amp uses GPT-6 Astra and Sign in with ChatGPT, is testing Ultrafast, and talks to Puck through GPT-Live-1, while agents plan and design as well as write code.
Searcher → Analyst → Writer → Editor · subagentic-20261014-202641
OpenAI published a customer story on October 14, 2026, about Amp, a North American technology startup building an AI agent that helps developers turn ambitious ideas into working software. The page is a startup case study, not a model launch. It describes Amp using GPT-6 Astra and Sign in with ChatGPT, talking with Puck through GPT-Live-1, and testing Ultrafast, with agents taking on planning and design as well as code. OpenAI lists Amp as an API customer.
Quinn Slack, Amp’s CEO and a member of its technical staff, says each model improvement is a chance to expand what the agent can do, from helping developers decide what to build to writing code and designing pages. Amp chooses and recommends models based on what works for its users. He says OpenAI models account for the majority of tokens used in Amp, and that OpenAI is transparent when safeguards are applied: “You can do all of your work in Amp with OpenAI models and be really happy.” Users, he adds, “really appreciate the consistency with which OpenAI models have been getting better and better.”
Tokens, then a longer leash
Token cost can become a constraint as builders give agents more work. Slack describes some Amp users consuming 10 billion tokens a week. If people watch every increment of usage, he believes, that may hold them back from finding out what an agent can do.
Amp launched Sign in with ChatGPT to help with that. Slack says it gives users access to more tokens and to OpenAI models, including GPT-6 Astra, which changes how willing they are to let an agent take on a larger task. “Sign in with ChatGPT means that people can have that feeling where they’re not counting every single token,” he says. “And once you actually let the agent go and have a longer leash, then you realize just how much it can do.” He says he sees people using the agent to plan, decide what to build, and continue working after they ship.
Writing and design, not only code
Amp wants agents on the work that goes into building and growing a product. Its own 20-person team uses them that way, including in marketing. “We don’t have someone devoted to it, but we have an agent devoted to it,” Slack says. The team relies on the agent to improve the experience for new users and to help more people discover Amp through search and other channels.
For that work, writing and design have to hold up. GPT-6 Astra helps the team produce documentation and design pages that meet Amp’s standards. “We have a very unique design style and writing style. We’re very picky about this, but we’ve been really happy with how GPT-6 Astra does,” Slack says.
His experiments with GPT-6 Astra continue at home. He gave it the plans for his house and used it to create a 3D model in Blender. He and his children then turn that model into games: a lazy river through the house, or chasing garlic around it, inspired by their grandmother’s dislike of garlic. “And that was not possible just a couple months ago,” he says.
Talking to Puck through GPT-Live-1
Amp recently announced Puck, a conversational assistant inside Amp that helps users launch and manage other agents. Users can ask Puck to start an agent to investigate a bug, coordinate several agents, or find a previous conversation about a technical problem.
Calling Puck, powered by GPT-Live-1, lets users talk to Puck in real time, including from a watch. Slack imagines people using it to keep agents working while they are away from their computers, even while out for a run. The page does not say voice itself launches those agents, and how widely that voice path is available is not stated.
Ultrafast is still a test
Amp has been testing Ultrafast mode with GPT-6 Astra. Slack says the greater speed changes how people use agents. Today, people often switch between several agent tasks while waiting for each to finish. Faster responses could let people juggle fewer agents, or give agents more time to check their own work before returning a result. “People have not seen a model that’s this good, that’s this fast,” he says. The page does not say Ultrafast is generally available.
For Amp, the next step is to bring more of the work before and after coding into the agent, so builders can carry an idea further without handling every step themselves. “There’s this unknown of what humans will create, and that’s what we all get to see. It’s cool to be playing a part in that,” Slack says.
Read the October 14 case study before treating any of this as a rollout. It is OpenAI’s account of what Slack says Amp is using, testing, and imagining.