How Amplitude Uses OpenAI Models to Ship AI-Driven Product Fixes in Days
Amplitude's proactive AI agent Wave leverages OpenAI's GPT-6.1 Sol, Luna, and Flare models to shorten product update cycles from a month to just 1–2 days.
Digital analytics provider Amplitude has introduced Wave, a proactive AI agent designed to automate the cycle of analyzing user telemetry, planning UX refinements, and dispatching code updates. By routing distinct analytical and generative tasks across specialized language and vision models, Amplitude reports that teams using Wave can collapse feature delivery timelines from roughly one month down to just one or two days. Approximately 80% of Wave’s underlying AI architecture relies on OpenAI models, including GPT-6.1 Sol, GPT-6 Luna, and GPT-Image-2.5 Flare.
Multi-Model Architecture for Automated Product Iteration
Wave functions by aggregating session replays, quantitative analytics, customer feedback, and underlying codebases to identify user friction. Principal AI Product Manager Jacob Newman explained that the goal is a unified platform where AI agents comb through massive data streams to surface high-value product updates. Once an issue is surfaced, Wave creates technical blueprints complete with wireframes and acceptance criteria before handing execution to automated coding assistants such as OpenAI Codex.
To evaluate Wave, Amplitude deployed the agent on its own homepage. Analysis of 100 session replays revealed that 80 visitors bypassed static AI suggestion cards to manually type queries, while only 12 clicked the preset prompts. Wave generated a technical proposal to dynamically personalize suggestion cards based on recent charts, dashboards, or referring URL context, while maintaining fallback defaults. Following human review and dispatch to a coding agent, an A/B test demonstrated a 160% increase in card-driven message dispatches after a single day compared to the control group. Senior Principal AI Engineer Eric Carlson noted that the result illustrates how automating routine, overlooked fixes can significantly increase feature adoption.
Strategic Benchmark Selection Across Reasoning and Cost
To balance operational expenses against output quality, Amplitude assigns specific tasks to specialized models based on quantitative benchmark testing. For complex investigations within its Opportunity Investigator pipeline, Amplitude selected GPT-6.1 Sol after it achieved an overall score of 4.06 out of 5—outperforming three alternative models that scored 3.62, 3.52, and 2.94—while completing investigations in a median time of 54 minutes. In separate testing for Wave's Data Science Agent, GPT-6.1 Sol scored 67.5 out of 100 compared to 69.3 for another leading model, but reduced average inference costs from $2.63 down to $1.28 per dispatch while excelling in tool-contract compliance and evidence grounding.
For lower-cost, high-frequency operations like code scanning and small evaluation checks, Amplitude routes requests to GPT-6 Luna. For visual mockups derived from session replay screenshots, GPT-Image-2.5 Flare scored 80.3 out of 100 across 15 test scenarios, outperforming a competitor model's score of 63.9 across metrics including text legibility and product fidelity. Amplitude CEO Spenser Skates highlighted that having access to both high-reasoning and lightweight models allowed the team to optimize quality, latency, and inference costs across the product. Skates also noted, "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 it means for developers
The architectural framework demonstrated by Amplitude highlights a maturing trend in AI software engineering: shifting from single-prompt interactions toward heterogeneous multi-agent systems. Rather than relying on a single large language model for all workloads, software teams are increasingly orchestrating workflows where smaller models manage telemetry scanning, specialized reasoning engines build execution plans, and vision models generate structural wireframes.
For developers building next-generation agentic workflows, accessing diverse model families is crucial for evaluating latency and cost trade-offs. Developers looking to implement multi-model architectures can test and deploy top AI models pay-per-token using a single API key through Apixo at https://apixoai.online.
Furthermore, Wave’s real-world beta deployment provides key insights into human-in-the-loop automation balance. While routine bug fixes are frequently routed straight to coding agents without manual intervention, larger product changes remain gated by human review. Despite these safeguards, velocity remains high: among customer-launched updates guided by Wave, 25% went live on the same day the opportunity was identified, 50% launched in under four days, and two-thirds shipped within a week. As Skates observed, the ultimate goal is to pioneer self-improving software development, enabling products to continuously iterate alongside engineering teams.
Source: How Amplitude ships product improvements in 1–2 days with OpenAI — OpenAI News. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
