Today we're releasing the first instance of self improving products with a customer: @PixAI_Official
Wave found that publishing artwork on PixAI's AI-image generation product was correlated with retention, but most new users never discovered the publish button. Wave suggested prompting users to publish immediately after generating an image. Now, when a user runs out of image generation credits, a tooltip points them to the Missions tab, where they can publish.
Results:
- Mission opens increased 76%
- Publish rate increased 10%
- Day 1 retention increased 3%
We're opening Wave to more teams who want to spend time with it. Let me know if you want to try it for your product!
every enterprise building agents should take note. masterclass in three steps:
- write great evals. make sure your first 100 are written by humans (ai slop evals are worthless)
- make your agent harness as model-agnostic as you can so you can effectively test every model release
- chart the pareto landscape like you see here & then model decisions become a really easy decision based on cost/quality/latency tradeoffs
btw this is the biggest practical challenge for the frontier labs going forward
i predict they will focus on "pacing the frontier" internally and instead turn their focus on incredible cost/performance for lower-tier models because 90%+ of global GPD requires nowhere near frontier-level intelligence
massive credit to the @Amplitude_HQ AI team for doing this better than most companies on earth