Amplitude has introduced Wave, a proactive product agent that the company says is roughly 80% powered by OpenAI models, including GPT‑6.1 Sol, GPT‑6 Luna, and GPT‑Image‑2.5 Flare. Wave’s role is to convert product analytics and user behavior into prioritized opportunities, technical plans, and measurable launches, coordinating execution with coding agents and tracking outcomes.
How Wave works and how models are assigned
Teams provide Wave with a product area and the goals or metrics they want to improve. Wave synthesizes Amplitude analytics, session replays, customer feedback, and code to:
- identify and prioritize opportunities;
- propose solutions and technical plans;
- coordinate implementation with people and coding agents (including Codex);
- measure results and feed learnings into subsequent iterations.
Different OpenAI models are assigned to different tasks. GPT‑6.1 Sol is used for the Opportunity Investigator and deeper code/data investigations; GPT‑6 Luna is used for frequent, lower‑cost tasks like scanning code and running small checks; and GPT‑Image‑2.5 Flare generates wireframes and mockups using screenshots from session replays as visual references. When an opportunity is ready, teams can send a technical plan and acceptance criteria from Wave directly to coding agents.
Example: personalizing suggestion cards on Amplitude’s homepage
Amplitude tested Wave on its own homepage, where generic suggestion cards were intended to help users start conversations with an AI agent. In a sample of 100 session replays, Wave found 80 users typed their own questions while only 12 selected a suggestion. Wave recommended personalizing the cards using recent charts, dashboards, or the page the user came from, illustrated the change, and produced a technical plan that preserved free typing and fell back to generic suggestions when personalization context was unavailable.
After a brief review the change was handed to a coding agent and A/B tested. Amplitude reported that messages sent via homepage suggestion cards rose 160% relative to control after one day. Eric Carlson, Senior Principal AI Engineer at Amplitude, described this as an example of how small, non‑groundbreaking changes can substantially increase feature adoption.
Model evaluations: quality, latency and cost
Amplitude evaluates quality, latency and cost for each task and assigns models accordingly. Deep investigations require strong reasoning and evidence‑backed conclusions; high‑volume repetitive checks require low latency and low cost.
In the Opportunity Investigator evaluation, GPT‑6.1 Sol scored 4.06 out of 5 overall; three alternative models scored 3.62, 3.52 and 2.94. GPT‑6.1 Sol led on all five investigation criteria and had the fastest median investigation time at 54 minutes.
In a separate Data Science Agent evaluation, GPT‑6.1 Sol scored 67.5 out of 100; another leading model scored 69.3. GPT‑6.1 Sol’s mean model cost per dispatch was $1.28 versus $2.63 for the other model. GPT‑6.1 Sol also scored higher on tool‑contract compliance, estimation and uncertainty, and evidence grounding in that evaluation.
For high‑volume tasks such as scanning code and running small evaluation checks, Amplitude uses GPT‑6 Luna to keep latency and cost low. Dividing work among specialist agents reserves more expensive reasoning for tasks that need it most.
For generating wireframes and mockups, OpenAI’s image model scored 80.3 out of 100 for solution wireframes versus 63.9 for another image generation model. The evaluation covered communication quality, product fidelity, instruction adherence, text and UI legibility, and visual coherence across 15 opportunities.
Customer usage and time to launch
Customers across industries are using Wave in beta to identify and ship product improvements. Wave provides a proposed fix, a technical plan and acceptance criteria; teams then choose the level of automation. Most teams send bug fixes straight to coding agents, while new features and larger changes get human review.
Amplitude reports that among customer‑launched experiments, feature flags and guides, one in four went live on the same day Wave surfaced the opportunity, half went live in under four days, and two‑thirds launched within a week.
Ambition and outlook
Amplitude says its ambition with OpenAI is to make self‑improving products the standard way companies build software. As OpenAI’s models improve at reasoning and coding, Amplitude expects teams to be able to delegate more investigation and implementation work to agents, accelerating the delivery of better products. The company is already seeing agents find and help deliver improvements in its own product while engineering teams work on other priorities.
Conclusion
Wave combines automated agents and human oversight while allocating different OpenAI models to tasks according to cost, latency and quality needs. Amplitude’s internal tests and homepage experiment suggest the approach can shorten time‑to‑ship and increase feature adoption, while monitoring model performance and costs for each class of task.
