Hey Atlassian dev!
We’re having a blast playing around with Rovo Agents and building some cool workflows. Most of the time, they work good, but we’ve noticed a few quirky patterns that we’d love to smooth out.
Here’s what we’ve seen:
- Occasional Workflow Hiccups: Sometimes (short percentage of attempts, but it is not suitable if we want to use it in automations), our agent pauses mid-workflow, says “I will now proceed to process the data,” and then does nothing
If we nudge it with “process it then,” it picks back up and finishes the job. What could be causing this? The possible complexity of the overall agent workflow?
- Inconsistent Issue Handling: The agent usually works great when we hit the “New” button to start fresh—it processes data like a pro. But when we ask it to handle data from one issue and then move to another (e.g., “process data from this issue” then “process data from another issue”), it sometimes skips parts of the workflow or misses some data. It’s like it’s not carrying the full context. Any ideas on why this happens?
- Workflow vs. Natural Abilities Mix-Up: We’re curious how the agent decides when to stick to its workflow versus using its natural smarts. For example, if we then ask it to summarize the processed data (the agent main goal is to return some data) or explain what the data’s about, it sometimes nails it but other times starts the workflow steps all over again. Is this tied to how we define the agent?
We’d love any tips or insights to help us get the most out of our Rovo Agents. Is it our prompt setup, or could context/caching be messing with things? Thanks in advance.