Amp is using OpenAI models across software development, documentation, design and conversational agent management. In an October 14 account published by OpenAI, Amp CEO Quinn Slack says those models account for the majority of tokens used in Amp. The company has also launched Sign in with ChatGPT, announced its Puck assistant and been testing Ultrafast mode with GPT‑6 Astra.
For developers, these features address different parts of the same workflow: getting access to models and tokens, assigning work to agents, and reducing the wait for results. The evidence describes Amp’s product direction and its team’s experience; it does not establish independently measured improvements in productivity, cost or software quality.
Sign in with ChatGPT addresses token constraints
Slack describes token costs as a constraint on how much work users give an agent. Some Amp users consume 10 billion tokens a week, he says. That figure describes some users, rather than typical consumption, and the account does not provide a distribution of usage or a corresponding bill.
Amp’s response is its Sign in with ChatGPT integration. According to Slack, it gives users access to more tokens and OpenAI models, including GPT‑6 Astra, making them more willing to delegate larger tasks. He describes users letting the agent help plan, decide what to build and continue working after a product ships.
The distinction matters: Slack’s argument concerns how perceived usage constraints affect delegation. It is not a quantified claim about savings. The supplied reporting does not specify token allowances, subscription eligibility, pricing or billing mechanics, so developers cannot use it to calculate what the integration would cost for their own workloads.
Hello Marvis AI Today’s coverage repeats Slack’s account of token usage and the integration. It also describes Amp’s model selection as case by case. OpenAI’s majority share of tokens therefore indicates substantial usage within Amp, without establishing an exclusive model relationship. The secondary coverage points back to OpenAI’s article and does not supply an independent usage audit.
GPT‑6 Astra supports work around the code
OpenAI’s account says Amp’s 20-person team uses agents for work that includes marketing, improving the new-user experience and helping people discover Amp through search and other channels. Slack says an agent handles that marketing work without a person devoted to it.
GPT‑6 Astra helps the team produce documentation and design pages, according to the same account. Slack reports satisfaction with its handling of Amp’s particular writing and design styles. These are concrete descriptions of internal use, but they do not include comparative quality scores or measurements of time saved.
The relevance is the breadth of the assignment. Amp describes agents participating in decisions before coding and in product work afterward. Documentation, page design and discovery give that ambition specific applications beyond generating software code.
Puck provides a conversational way to manage agents
Puck is a conversational AI assistant built into Amp to help users launch and manage other agents, according to OpenAI. The documented examples include asking it to start an agent to investigate a bug, coordinate several agents or find a previous conversation about a technical problem.
Calling Puck adds real-time conversation through GPT‑Live‑1, including access from a watch. This gives the voice feature a distinct role: users can talk to an assistant that helps manage agent work. The source does not say that GPT‑Live‑1 performs all the underlying development tasks.
Slack imagines users keeping agents working while away from their computers, including during a run. That is his proposed use case; the account does not establish how widely people already use Puck that way. It also does not detail supported watch models, permissions or how consequential actions are approved.
Ultrafast remains a test, with benefits still unmeasured
Amp has been testing Ultrafast mode with GPT‑6 Astra. Slack says greater speed changes how people use agents. OpenAI’s account suggests faster responses could let users juggle fewer concurrent agent tasks, or leave agents more time to check their work before returning a result.
Those are potential workflow benefits. The supplied evidence gives no latency figures, throughput measurements or controlled comparison, and it does not establish general availability or access terms for the mode.
Taken together, Amp’s integrations describe a push to make agents useful across more of a product’s development and operation. For a developer evaluating that approach, the unresolved questions are practical: what access costs, which controls govern delegated work, and whether faster interactions produce better completed results. The reported use cases explain Amp’s direction, while leaving those outcomes to be demonstrated.
