OpenAI published a customer story on 12 November 2026 describing how Amplitude, a product analytics company, uses its models inside Wave, a "proactive product agent" that looks for ways to improve a product and then helps ship them.
The structure of the claim is straightforward. Wave is described as roughly 80% powered by OpenAI models: GPT-6.1 Sol for deep investigation of code and data, GPT-6 Luna for frequent lower-cost work such as scanning code, and GPT-Image-2.5 Flare for generating mockups from session-replay screenshots. When an opportunity is ready, teams can push it to coding agents including Codex with a technical plan and acceptance criteria.
The headline number is time. "Amplitude estimates that Wave can help teams ship product improvements in 1-2 days, down from roughly a month." That is the company's own estimate, and the source presents it as an estimate rather than a measurement.
The worked example is Amplitude's own homepage. Generic suggestion cards were meant to start conversations with an AI agent. Wave examined 100 session replays and found that 80 users typed their own questions while only 12 picked a suggestion. Wave proposed personalising the cards using recent charts, dashboards, or the referring page, produced a technical plan that preserved free typing, and the change went to a coding agent and then to an A/B test. Amplitude reports that messages sent through the cards rose 160% relative to control after one day.
Eric Carlson, a senior principal AI engineer at Amplitude, framed that result less grandly than the headline invites: "It is a great example of how non-groundbreaking things that just get ignored can increase the adoption of features in unexpected ways."
The piece also publishes model evaluation figures, and they are not uniformly flattering. On the Opportunity Investigator task, GPT-6.1 Sol scored 4.06 out of 5 against 3.62, 3.52 and 2.94, and had the fastest median investigation time at 54 minutes. On a separate Data Science Agent evaluation, GPT-6.1 Sol scored 67.5 out of 100, below the 69.3 recorded for another leading model, while carrying a mean model cost of $1.28 against $2.63 per dispatch on that task.
Spenser Skates, Amplitude's chief executive, is quoted on the division of labour: "AI can now find useful work we never thought to assign it. OpenAI's models can investigate a problem, write and run code, and use the results to decide what to try next."
What the document does not contain matters as much as what it does. There is no independent audit of the shipping-time estimate. The alternative models in the comparison are not named. No sample size is published for the Opportunity Investigator or Data Science Agent evaluations, and the full evaluation harness is not published. The 160% figure rests on a single A/B test run by Amplitude on its own homepage.
Read as a vendor case study, the piece is coherent and, on cost, more candid than most. Read as evidence that AI agents reliably shorten product cycles across the industry, it is a sample of one, measured by the party selling the tools.