Amplitude Uses OpenAI-Powered Wave to Speed Product Changes

Amplitude says its proactive product agent, Wave, can help teams ship product improvements in one to two days, compared with roughly a month. Built roughly 80% on OpenAI models, Wave combines product data with implementation workflows and can pass technical plans to coding agents, including Codex.
Amplitude tested the system on its own homepage, where suggestion cards were intended to help users start conversations with an AI agent. The company reported a 160% lift in messages sent through those cards versus the control group after one day of an A/B test.
From product signals to implementation plans
Teams give Wave a product area plus the goals and metrics they want to improve. The agent brings together Amplitude analytics, session replays, customer feedback and code to identify and prioritize opportunities, propose solutions, coordinate implementation, and track the results.
For the homepage experiment, Wave reviewed a sample of 100 session replays. Amplitude found that 80 users typed their own questions, while 12 chose a suggestion card. Wave proposed personalizing cards from a user’s recent charts, dashboards or referring page, while retaining free typing and generic suggestions where no personalization context was available.
The agent also produced a technical plan and acceptance criteria before the team reviewed the work and handed it to a coding agent. That workflow follows the broader pattern in which OpenAI models in agents beyond code tasks illustrates how OpenAI models are being integrated into agents that extend beyond pure code generation.
Different models for different tasks
Amplitude assigns models according to the task’s required reasoning depth, speed and inference cost. GPT-6.1 Sol supports Wave’s opportunity manager and deeper code and data investigations, while GPT-6 Luna handles frequent work such as code scanning and smaller evaluation checks. GPT-Image-2.5 Flare generates product mockups using screenshots from session replays as visual references.
In Amplitude’s Opportunity Investigator evaluation, GPT-6.1 Sol scored 4.06 out of 5 overall, ahead of three alternative models that scored 3.62, 3.52 and 2.94. It led across corroborating findings, avoiding false positives, incorporating evaluator feedback, recovering from tool failures and aligning claims with evidence. Its median investigation time was 54 minutes.
In a separate Data Science Agent evaluation, GPT-6.1 Sol scored 67.5 out of 100, compared with 69.3 for another leading model, while its mean model cost was $1.28 versus $2.63 per dispatch. For solution wireframes, GPT-Image-2.5 Flare scored 80.3 out of 100, compared with 63.9 for another image model across 15 opportunities.
Human control remains part of the workflow
Amplitude says customers using Wave in beta can choose their level of automation. Teams commonly send bug fixes directly to coding agents, while larger feature changes and new bets receive human review. Among customer-launched experiments, flags and guides, Amplitude says one in four went live on the day Wave surfaced the opportunity, half launched in under four days, and two in three within a week.
The business implication is that teams considering agent-led product work need explicit success metrics, acceptance criteria and human approval rules, then must measure whether a released change improves the intended user outcome.

