← /editions
#154

Amplitude's Wave turns the product loop into an agent loop

// Amplitude's proactive product agent, Wave, identifies what to build, ships it with coding agents, and measures the impact, cutting product improvements to 1–2 days.

$ today_in_10_seconds

Amplitude's product agent, Wave, doesn't wait to be asked. It finds what to build, hands the work to coding agents, and measures what happened, cutting product improvements to 1–2 days. The interesting part isn't the speed. It's that the agent owns the whole loop, not just the code.

$ lead_story

Amplitude's Wave turns the product loop into an agent loop

Amplitude has shipped Wave, a proactive product agent built mainly on OpenAI models. It spots what to build from product data, helps teams ship the change with coding agents, and then measures the impact. Most "agentic" dev tools today sit in the middle of that chain and wait for a human to write the ticket. Wave starts before the ticket and keeps going after the merge. For builders, the lesson is that a durable moat is the data source that tells the agent what to do and whether it worked, not the code generation, which keeps getting cheaper.

// why_it_matters: When an agent picks the work and grades its own results, the human job shifts from writing tickets to setting goals and guardrails.

read more →

$ bigger_picture

The trend behind the headlines

Coding agents made the "build" step cheap, which exposed the slow parts on either side: deciding what's worth building and proving it mattered. Companies that sit on behavioral data, like analytics, observability, and CRM vendors, are now in a position to wrap agents around both ends. Expect more "proactive" agents that open the PR before anyone asks. The open question is trust: an agent that both proposes and measures its own changes needs outside checks, or it will optimize for whatever metric is easiest to move.

$ tool_of_the_day

OpenAI Agents SDK

Wave runs mostly on OpenAI models, and the Agents SDK is the lightweight way to build a similar multi-step loop yourself: tools, handoffs between specialized agents, guardrails, and tracing built in. Tracing is the feature to care about if your agent acts on its own; you'll want a record of every decision it made.

explore OpenAI Agents SDK →

$ subscribe --free

Get the next edition in your inbox

// DAIly Agent covers agentic AI three times a week

// free, no spam, exit anytime