Launching today
LayerLens Environments
Test AI agents against simulated business services
1 follower
Test AI agents against simulated business services
1 follower
For teams building AI agents: test workflows in simulated business services, add missing cases with Synthetic Data, and evaluate outputs and actions with reference answers and model-based rubrics. Inspect the evidence before deciding what to change.
Hi Product Hunt, Jake from LayerLens here.
In our HubSpot test this week, 7 saved agent runs answered questions in a simulated HubSpot account with fictional records. All 7 passed the exact-value grade. On one question, "how many deals are open?", 3 runs gave the right answer, 8, by 3 different routes. One of them read a single deal and repeated the total the search reported. It passed, and the grade alone could not tell you that.
That is why we built LayerLens Environments. You give an agent a task inside a simulated business service, it works through the same kinds of tools it would use on the real service, and you get the result plus a trace of every call it made.
The saved traces for 2 of those runs are public: https://huggingface.co/datasets/LayerLens/hubspot-open-deal-count-test
What would you need to see in a trace before trusting an agent with your own records?
Jake