Industry

Amplitude Cuts Product Development Cycles to 1–2 Days Using OpenAI Models

Amplitude's Wave agent, powered roughly 80% by OpenAI's GPT-6 family, automates product insights-to-shipped-code workflows, reducing shipping timelines from ~1 month.

Last verified:

Wave: From Data to Shipped Code in Days

Amplitude, a product analytics platform, has deployed Wave—a proactive agent system designed to compress the entire product improvement cycle. According to the OpenAI Blog, Wave automates the journey from identifying user behavior patterns to deploying and measuring code changes. The agent is approximately 80% powered by OpenAI’s latest models, turning raw product data into prioritized improvement candidates and ultimately coordinated implementation.

This represents a fundamental shift in how teams approach product iteration. Rather than engineering-led roadmap planning followed by months of development, Wave ingests product analytics, session replays, and customer feedback to surface opportunities, propose solutions, and orchestrate handoff to coding agents for implementation. The reduction in cycle time—from roughly one month to an estimated 1–2 days—reflects the elimination of bottlenecks in opportunity discovery and cross-functional alignment.

Task-Specific Model Deployment

Amplitude’s architecture distributes work across three OpenAI models based on computational and financial efficiency. GPT-6.1 Sol handles the highest-complexity operations, serving as the engine for the opportunity manager and conducting detailed investigations across codebases and product metrics. GPT-6 Luna takes on frequent, routine workloads—scanning code repositories and executing small evaluation checks—where lower cost per inference is prioritized over reasoning depth. GPT-Image-2.5 Flare generates visual prototypes by using session replay screenshots as anchors for product mockup generation.

This tiering strategy allows Amplitude to optimize for both capability and cost. The company does not run every task through its most capable model; instead, it routes simpler operations to faster, cheaper alternatives, reserving advanced reasoning for discovery and prioritization steps where it provides the most value.

Real-World Impact: The Homepage Suggestion Card Test

Amplitude tested Wave internally on its own product homepage, where generic suggestion cards were intended to prompt users to initiate conversations with an AI assistant. The team ran Wave on session replay data and discovered that users frequently bypassed the generic prompts—instead typing their own questions. According to the OpenAI Blog, Wave proposed personalizing the suggestion cards by pulling context from recent charts, dashboards, or the user’s current page, while preserving the option for free-form typing and falling back to generic suggestions when context was unavailable.

The personalization proposal included both a visual illustration and a technical implementation plan with acceptance criteria. After brief review, the work was handed to a coding agent for implementation and A/B testing. The result: messages sent through the suggestion cards increased 160% relative to control within one day of deployment.

Why This Matters

The Wave deployment signals a major acceleration in the latency between insight and shipped improvement, a constraint that has historically required weeks of product design and engineering cycles. Teams constrained by bandwidth or lacking dedicated product analytics expertise may now use AI agents as force multipliers, reducing the friction between identifying a problem and testing a solution.

The reliance on OpenAI’s model family—particularly the routing of lower-cost operations to GPT-6 Luna—suggests that specialized open-weights models or smaller proprietary alternatives may struggle to compete in workflows where cost-per-token becomes the marginal optimization. For product teams evaluating internal agents versus managed platforms, the question shifts from “can AI help?” to “can our cost structure support continuous agent-driven iteration?”

Frequently Asked Questions

What OpenAI models power Amplitude's Wave agent?

Wave is approximately 80% powered by OpenAI models: GPT-6.1 Sol for deep reasoning and opportunity prioritization, GPT-6 Luna for frequent lower-cost tasks like code scanning and evaluation checks, and GPT-Image-2.5 Flare for generating product mockups from session replays.

How much faster does Wave make product shipping?

According to the OpenAI Blog, Amplitude estimates Wave can help teams ship product improvements in 1–2 days, down from roughly one month—a substantial compression of the development cycle.

What was the result of Wave's first internal test?

When applied to Amplitude's homepage suggestion cards, Wave proposed a personalization strategy that increased messages sent through suggestion cards by 160% relative to control after one day.

#OpenAI #product-development #AI-agents #code-generation #Amplitude