Amp Launches Puck and Tests Ultrafast with OpenAI

Amp is leveraging OpenAI models to extend the capabilities of its agents beyond coding. The company has introduced Puck, a conversational assistant that can be summoned in real-time, is testing an Ultrafast mode with GPT-6 Astra, and offers Sign in with ChatGPT to provide access to more tokens. Its team is already using these agents for writing, design, and marketing, aiming to integrate more steps before and after development.

Puck orchestrates agents and can be invoked in real-time from a watch

Amp has unveiled Puck, an integrated assistant that helps launch and manage other agents. Users can ask it to open a bug report, coordinate multiple agents, or retrieve a previous technical conversation. With Calling Puck, powered by GPT-Live-1, the exchange occurs in real-time, even from a watch. Quinn Slack envisions that users could keep their agents active even when away from their computers, including while running.

An Ultrafast mode tested with GPT-6 Astra and expected effects on usage

Amp has tested the Ultrafast mode with GPT-6 Astra. According to Quinn Slack, the speed gain alters how agents are used. Currently, it is common for users to switch from one agent task to another while waiting each time, whereas faster responses could either reduce the number of agents to manage or allow more time for each agent to control its own output before delivering it. He believes that users have never experienced a model that is both this fast and this effective.

Sign in with ChatGPT aims to ease token constraints

Quinn Slack notes that the cost in tokens can become a constraint when more tasks are assigned to agents, with some users reaching 10 billion tokens per week. He thinks that overly strict tracking of each usage increment can limit the exploration of an agent's capabilities. To address this, Amp has launched Sign in with ChatGPT, which provides access to more tokens and OpenAI models, including GPT-6 Astra. He believes this option changes users' willingness to assign more significant tasks to an agent.

Writing, design, and marketing: GPT-6 Astra already equips a team of 20

Amp's team, consisting of 20 people, uses agents for various functions, including marketing. There is no dedicated position, but an agent is assigned to this task to improve the onboarding of new users and facilitate discovery through search and other channels. High standards in writing and design are applied, and GPT-6 Astra is used to produce documentation and design pages that meet the company's standards. Quinn Slack also mentions a personal use: he provided the plans of his house to GPT-6 Astra to create a 3D model in Blender, which he then turns into games with his children, an experience he deemed impossible just a few months ago.

OpenAI is predominant at Amp, aiming for agents that cover more than just code

Amp recommends models based on their performance with users and notes that those from OpenAI account for the majority of tokens used. Quinn Slack highlights OpenAI's transparency and finds their models easy to use, stating that they are sufficient to accomplish all the work at Amp, with a perceived constant improvement. The company is developing an agent to transform ideas into software, and according to Quinn Slack, the expansion of capabilities follows the evolution of models, from decision support to design. Uses include planning, feature selection, and continuing work after production deployment. Amp is now aiming for agents involved in all aspects of product building and growth, with the next step being the integration of tasks upstream and downstream of coding so that creators can advance an idea without managing every step. Quinn Slack finally emphasizes the uncertainty about what humans will create and expresses his happiness to contribute to this process.