Amp is broadening the role of its AI agents so they help developers with more than just writing code. The company integrates several OpenAI models — including GPT‑6 Astra, Ultrafast, GPT‑Live‑1 and Sign in with ChatGPT — to enable agents to assist with planning, documentation, page design, coordination and other non‑coding tasks.
Why OpenAI models?
Amp selects and recommends models based on what works best for users, and OpenAI models account for the majority of tokens Amp consumes. Quinn Slack, Amp’s CEO, says OpenAI’s transparency and the presence of safeguards make those models straightforward to use: users understand when limits are applied, which helps trust and adoption.
Token costs and Sign in with ChatGPT
As agents take on more work, token costs can become a constraint. Amp has observed some users consuming as much as 10 billion tokens per week. Slack argues that tight monitoring of every token can prevent people from discovering the full capabilities of an agent.
To address that, Amp launched a Sign in with ChatGPT integration. According to Slack, it gives Amp users access to more tokens and OpenAI models, including GPT‑6 Astra, and changes how willing users are to let an agent handle larger, longer tasks — because they are not counting every token.
Agents across product workflows
Amp aims for agents to assist with all the work involved in building and growing a product. Its own roughly 20‑person team uses agents across tasks, including marketing: rather than a dedicated marketing hire, the company uses a dedicated agent for that role.
The team relies on agents to improve onboarding, help with discoverability via search and other channels, and produce higher quality writing and design. GPT‑6 Astra supports the team in producing documentation and design pages that meet Amp’s standards.
Puck and real‑time interaction
Amp recently announced Puck, a conversational AI assistant embedded in Amp that helps users launch and manage other agents. Users can ask Puck to start an agent to investigate a bug, coordinate multiple agents, or locate a past conversation about a technical issue.
Calling Puck, powered by GPT‑Live‑1, enables real‑time voice interaction, including from a watch. Slack envisions people using it to keep agents working while away from their computers, for example while out for a run.
Speed changes usage patterns
Amp has been testing Ultrafast mode with GPT‑6 Astra; Slack says higher speed alters how people use agents. Today, users often switch among several agent tasks while waiting for each to complete. Faster responses could reduce the need to juggle many agents or give agents more time to self‑check before returning results.
Practical examples and next steps
Slack’s own experiments illustrate new possibilities: he provided GPT‑6 Astra with plans for his house and used it to create a 3D model for Blender, which he and his children turned into games — for example adding a lazy river through the house or a playful garlic‑chase inspired by family stories. He notes that similar creative uses were not possible just a few months ago.
Amp’s next goal is to move more of the pre‑ and post‑coding work into the agent, so builders can carry ideas further without manually handling every step. Slack says the unpredictable results of what humans will create with these tools are part of the appeal, and Amp aims to play a role in that exploration.
Key facts
- OpenAI models used: GPT‑6 Astra, Ultrafast, GPT‑Live‑1, Sign in with ChatGPT integration
- Token consumption observed for some users: up to 10 billion tokens per week
- Amp team size: about 20 people
- New features: Puck (conversational assistant), Calling Puck (real‑time interaction via GPT‑Live‑1)
- Use cases: planning, decision making, documentation, design, real‑time agent management, creative 3D modelling
