If you’ve worked with voice agents, this might sound familiar:
You run a big batch of tests in your simulator.
Some calls fail in odd ways.
You open the workflow graph and start replaying calls, node by node, trying to find the moment the agent “went off script.”
PMs and engineers using SIMULATE were doing this all the time.
The question was always the same:
“Where exactly did this agent’s path diverge from what we designed?”
The process was slow, manual, and repetitive but also extremely valuable.
So we shipped it as a feature: Flow Analysis.
With Flow Analysis, each test run gives you:
- Full path trace: the exact route an agent took through your workflow
- Divergence point: the node where it broke away from the expected path
- Conversation context: how the rest of the interaction unfolded from that point
This turns debugging from “scrub and guess” into a clear, visual diff between expected vs actual behavior.
Instead of hunting through graphs, you can focus on:
- Fixing misrouted branches
- Adjusting conditions or thresholds
- Improving prompts and error handling where it actually matters
If you’re building or testing voice agents and still doing manual graph forensics, Flow Analysis might save you a lot of time.
🔗 More details: https://shorturl.at/Ia2tG
Top comments (0)