Issue Description
Agent does not take follow up questions or keep context in memory. Chat Assistant is able to take follow up questions, but Agent cannot.
Root Cause (if known)
From Glean support feedback, this seems to be currently for both Chat and input form triggered agents as designed. Any follow up questions after initial responses re-runs entire agent workflow, not responding to instructions on prior output modification only.
Proposed Solution/Enhancement
Like with Chat Assistant, Agents should have capability to answer follow-up questions to modify output of prior responses beyond having to re-run the full agent flow again. The weird thing is that we think in Chat triggered agents it actually seems to partly support this currently as we did get some mixed results on that modifying specific output to let's say "Make it a table" on Strenghts of software vendor document for example in prior output. While the input form based triggered agent, did not seem to do so well on those follow up outputs. Although both agents show they re-ran the full workflow. The table only showed up with strengths and not weaknesses of the document we used, so somehow it did retain some prior question/output memory?
This feature could be made into a configuration setting at Agent level or RESPONSE Step level in our opinion to allow or not allow this option.
Impact
Users expect especially in chat triggered Agents but also normal input form ones they prompt to have options to modify outputs to their liking especially being used to copilot chat agents and chatgpt like interfaces. This context memory follow up should be a baseline feature for agent interactions by users to customize output further after initial responses, till they have output results that meet their needs.
