2026-11-12 · ← News
Amplitude cuts product changes to 1 or 2 days, but code is no longer the bottleneck
Amplitude says its Wave agent can move some product changes from signal to deployment in 1 or 2 days. Its own numbers expose the harder limit: only about 11% of proposed changes reach production in a mature codebase.
The image could not be loaded.
Amplitude has introduced Wave, an agent that finds problems in product data, drafts specifications, routes work to coding agents and measures results after deployment. An OpenAI customer story says the company can ship some changes in 1 or 2 days this way.
Wave joins product data, code and measurement in one loop
Wave reads product analytics, feedback, session replay, error logs, agent traces and experiment results. It turns those signals into a problem with evidence, proposes a solution and routes the work to a person or coding agent according to risk. A separate review agent checks the change against the specification, runs it in a sandbox and prepares a recording for human review.
Amplitude says the first version of its recommendations earned only about 5% positive feedback. Internal iteration raised that figure above 70%. The company also says it processes almost 80 billion events a day and more than 2.5 trillion a month, giving Wave a broad base for finding product problems.
Faster coding moves the argument to what deserves to be built
A coding agent can execute a precise specification quickly. Choosing a change that helps users and moves the right metric is much harder. Wave therefore operates one level above a conventional bug bot by connecting a technical issue with user behavior, a business objective and an experiment outcome.
That changes the purpose of review for product teams. A person must judge evidence, priority and strategy as well as code quality. If that step is weak, a fast agent merely accelerates the production of well written distractions.
Agents sprint in fresh repositories and stop at the mature product gate
Amplitude describes a sharp gap between a demo and production. On a new, low risk project, an agent created 101 features in one week. In the company’s mature codebase, however, only about 11% of proposals reach the finish line even though the team would like to pursue almost 50%. Weak citations, stale context, repository discovery and demanding human review are the blockers.
Wave is also in a closed beta for a small group of design partners. The figures come from Amplitude and are not yet an independent measure of general product performance. The company says it does not train models on customer data, but buyers will still need precise processing and access terms.
Production survival rate will decide whether the loop works
The useful metric will not be opportunities found or pull requests opened. Teams should track the share of proposals deployed, human verification time, incorrect citations and actual movement in retention or revenue metrics.
Wave succeeds if it lifts today’s roughly 11% without increasing production defects or making product managers spend the day correcting its judgment. Only then do 1 or 2 days become an operating model rather than a customer story headline.
Lilith's verdict
Wave can bring a pull request to the door in 2 days. A person still sits inside and must decide whether the change helps a customer or merely keeps the review queue busy.