Editorial illustration for Amplitude uses OpenAI's Codex to ship product improvements in 1–2 days
Amplitude Ships Product Updates in Days Using AI
Amplitude uses OpenAI's Codex to ship product improvements in 1–2 days
Amplitude built its name on tracking how people use software. Now it's using that data to decide what to build next, and shipping the result in a day or two instead of a month. The company's proactive product agent, called Wave, runs on OpenAI models for roughly 80% of its work, pulling from GPT-6.1 Sol, GPT-6 Luna, and GPT-Image-2.5 Flare. Instead of waiting for a product team to spot a trend in a dashboard, Wave scans usage data, session replays, and customer feedback on its own, flags what's worth fixing, and drafts an implementation plan that gets handed to coding agents like Codex to build.
The numbers Amplitude is citing are specific: a 160% lift in agent engagement after one day on its own homepage, where Wave personalized suggestion cards and pushed up card-driven sends. On the cost side, GPT-6.1 Sol came in 51% cheaper per dispatch in Amplitude's Data Science Agent evaluation, and ran $1.34 per task in its Opportunity Investigator tests. The pitch is that Wave doesn't just surface problems, it closes the loop from insight to shipped code to measured outcome.
Amplitude estimates that Wave can help teams ship product improvements in 1–2 days, down from roughly a month. On Amplitude’s own homepage, Wave helped personalize suggestion cards, driving a 160% lift in card-driven sends.
Why this matters Amplitude's pitch here is less about Codex writing code and more about collapsing the distance between "we noticed something in the data" and "we shipped a fix," down to 1-2 days. For teams building AI products, that loop speed matters more than any single benchmark. Wave handing off a technical plan and acceptance criteria to a coding agent is a real workflow pattern worth watching, not just a demo.
That said, the numbers deserve scrutiny before anyone copies this. A 160% lift in agent engagement "after one day" is an awfully short window to call a win. And the cost figures, 51% lower mean model cost per dispatch, $1.34 for GPT-6.1 Sol, come from Amplitude's own Data Science Agent evaluation, with no detail on baseline model, task volume, or how "dispatch" is defined. Vendor-reported efficiency gains in isolation tell us little.
What's actually interesting is the structural bet: letting one agent decide what to build and another execute it, with humans mainly setting guardrails. If that holds up with messier products and longer timeframes, it's a meaningful shift in how product teams operate. Right now, it's one customer's early numbers.
Related Reading
Browse all OpenAI news →
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google to End Free Access to Gemini AI for Non-Subscribers
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic OpenAI AGI chief takes leave as company buys viral talk show TBPN today
Business & Startups
Related Topic OpenAI's GPT Transcribe Cuts Error Rate to 3.31%, Improving on GPT-4o
Open Source
From Archives Anthropic launches Claude Meetings, putting AI into video calls
From Archives Google's new Gemini model becomes AI Mode default for hardest questions
From Archives OpenAI Launches India API Inference Residency for Data Storage and Processing
Business & Startups
From Archives TSMC board approves USD 500M OpenAI and USD 500M Anthropic investments
Business & Startups
Common Questions Answered
What is Wave and how does it help Amplitude ship product improvements faster?
Wave is Amplitude's proactive product agent that runs on OpenAI models for approximately 80% of its work, utilizing GPT-6.1 Sol, GPT-6 Luna, and GPT-Image-2.5 Flare. Instead of waiting for product teams to manually spot trends, Wave automatically scans usage data, session replays, and customer feedback to identify improvement opportunities and generate technical plans with acceptance criteria. This automation enables Amplitude to ship product improvements in just 1-2 days compared to the previous timeline of roughly one month.
What specific results did Wave achieve when personalizing suggestion cards on Amplitude's homepage?
Wave helped personalize suggestion cards on Amplitude's own homepage, which resulted in a 160% lift in card-driven sends. This real-world application demonstrates the practical impact of using AI-driven product optimization to enhance user engagement metrics.
Which OpenAI models power Wave and what percentage of its work do they handle?
Wave runs on OpenAI models for roughly 80% of its work, specifically pulling from GPT-6.1 Sol, GPT-6 Luna, and GPT-Image-2.5 Flare. These models work together to analyze data, generate insights, and create technical plans for product improvements.
How does Wave's workflow pattern differ from traditional product development processes?
Wave collapses the distance between identifying a data trend and shipping a fix down to just 1-2 days by automating the discovery and planning phases. Rather than requiring manual review of dashboards and separate technical planning, Wave hands off a complete technical plan and acceptance criteria directly to a coding agent, creating a continuous loop that prioritizes speed over traditional development cycles.
Further Reading
- How Amplitude ships product improvements in 1–2 days with OpenAI - OpenAI
- How Amplitude ships product improvements in 1–2 days with OpenAI - OpenAI
- Amplitude's Wave agent decides what to build, ships the code, and measures impact, all on OpenAI models - Nota Tech Guy
- Introducing the Codex app - OpenAI
