The customer need
Product teams have plenty of signals. The difficult, time-consuming work is connecting them: understanding where people struggle, deciding which problems matter, and building confidence to act.
Wave is great at surfacing useful problems, but dense proposals, product images that were an approximation of the real thing, and a single proposed solution made them hard to assess and refine.
We needed to make it easier to answer three questions: Do I understand this? Do I believe it? Is it worth doing?
Why it mattered
Customers needed to spend less time assembling evidence and more time driving impact against their goals. For Amplitude, delivering that value was essential to repeat use.
My role
I joined Wave in July as the team’s only designer during its 0→1 development. We partnered with 49 customers to help shape and build Wave. The early, AI-designed experience had accumulated features and configuration that obscured the core workflow. I led the work to cut that bloat and shape opportunity review, solution refinement, and handoff.
I partnered with Product, Engineering, the agent team, and Growth toward a November launch. As the strategy evolved, we balanced improvements to opportunity quality with design interventions, distinguishing when a recommendation needed stronger evidence or reasoning from when clearer presentation or interaction could help teams evaluate and refine it.
The decisions that shaped the experience
Open a decision to read the thinking behind it.
01Bring the problem and solution back together
Three tabs reflected a staged workflow but made people assemble the idea themselves. Customer calls and prototypes showed that seeing the problem and proposed change together made the opportunity easier to understand.
On customer calls, people often skipped the problem and evidence and went straight to the visuals to understand the opportunity. I brought the problem, evidence, solution, and acceptance criteria into one scannable page, pairing the visuals so people could grasp the change before reading the plan.
Earlier views asked PMs to evaluate too much evidence at once, making it hard to separate Wave’s conclusion from its supporting sources. I paired a concise problem statement with the three most relevant sources inline, keeping the full evidence report available for deeper investigation. A PM could understand the opportunity first, then follow a citation to check an important or questionable claim. The report carried the fuller story: what happened, why, and what it costs.
02Give people room to shape the solution
Wave could identify a promising problem, but a single generated solution felt too final when the design might need iteration or a different approach. This particularly limited design-led work. Users also missed that they could shape the solution in Global Chat.
We explored several approaches, including opportunity templates tailored to the problem. A simple frontend change could show one proposed solution. A more complex opportunity spanning a flow or multiple screens could offer alternative directions and a downloadable design kit for Claude Design or another tool of the team’s choice.
After testing prototypes with customers, we deferred the broader solution-shaping UI until we had more feedback about which workflows were most useful. Our near-term decision was to make it clearer that teams could already iterate on a solution through Global Chat, a capability that hadn’t been obvious.
03Shift the focus toward teams’ existing workflows
We partnered with Growth on a sprint through the full Wave loop, from opportunity review to implementation and measurement. They found useful opportunities but struggled with execution, reinforcing customer feedback.
We deprioritized execution within Wave while continuing to offer it, and focused on Linear ticket creation. The Linear/Jira handoff was improved to carry the full opportunity context and wireframes into teams’ existing workflows.
We gave Linear/Jira tickets and their statuses precedence over Wave’s own status system, cutting duplicate status tracking. That decision shaped the opportunity table for each product area: teams could follow the work through the ticket statuses they already used.
What changed and what I learned
I learned to design AI as a collaborator in a team’s decision-making, not as the author of a single, definitive answer. Teams need to be able to question the evidence, explore alternatives, and shape the recommendation before acting on it.
- Trust depends on showing the work. Evidence must be easy to inspect, metrics understandable, and visuals recognizable. Presentation helps people judge the recommendation.
- The handoff is part of the experience. Evidence and intent need to survive the move into other tools, with a record of what actually ships.
- Setting the product boundary is design work. Cutting execution scope focused Wave on work it could make more useful: identifying problems and preparing plans that fit existing workflows.
- Impact is the measure of success. Growth shipped five opportunities, but verification and measurement still caused friction. I learned to judge the full effort required to act on a recommendation and learn whether it improved the product against the team’s goals.