Amplitude Wave Product Agent Integration with OpenAI Models

Overview of Amplitude Wave

Amplitude has introduced Wave, a proactive product agent designed to help teams identify, prioritize, and implement product improvements. Wave leverages a suite of OpenAI models—specifically GPT-6.1 Sol, GPT-6 Luna, and GPT-Image-2.5 Flare—to analyze product data, session replays, and customer feedback. By automating the transition from insight to implementation, Amplitude reports that teams can reduce the time required to ship improvements from approximately one month to 1–2 days.

Model Specialization and Task Allocation

Amplitude employs different OpenAI models based on the specific requirements of the task, balancing reasoning capabilities against latency and cost.

  • GPT-6.1 Sol: Utilized for high-reasoning tasks such as the Opportunity Investigator and deep investigations into code and data. In internal evaluations, this model outperformed competitors in investigation criteria and demonstrated superior tool-contract compliance and evidence grounding.
  • GPT-6 Luna: Deployed for high-volume, lower-cost tasks including scanning code and performing frequent, small evaluation checks.
  • GPT-Image-2.5 Flare: Used for generating product mockups and wireframes, utilizing screenshots from session replays as visual references.

Performance and Implementation

Wave functions by aggregating product analytics and session data to propose solutions, which are then converted into technical plans and acceptance criteria. These plans can be dispatched directly to coding agents, such as Codex, for execution.

Amplitude demonstrated the efficacy of this workflow on its own homepage. By using Wave to identify that users were bypassing generic suggestion cards, the team implemented personalized suggestions based on recent user activity. This change resulted in a 160% increase in card-driven message sends. Across broader customer use cases, Amplitude reports that two out of three launched experiments, flags, or guides go live within a week of Wave surfacing the opportunity.

Human-in-the-Loop Control

While Wave automates the discovery and planning phases, Amplitude maintains human oversight for the final implementation. Teams retain control over the level of automation applied to specific tasks. While routine bug fixes may be sent directly to coding agents, significant new features and strategic updates typically undergo human review before deployment. This approach allows organizations to balance the speed of agent-led discovery with the necessity of human judgment for product-critical changes.

Sources