Q&A generated from Webex transcript and chat
Q: Is it possible to work with real time data being managed by an organization within a secured cloud?
A: Absolutely. Whether it’s in our cloud or whether it’s in a different cloud, we have connectors that give you access to all the standard data lakes and data warehouses in the market and you’re able to pull them in as data sets, add your semantic information and start using them with your agents.
Q: In terms of traceability, can agents surface the specific SQL or data transformations they performed to deliver final reports or responses?
A: Absolutely. It’s important to be transparent with the end users with the SQL that we generate. When you open up the explanation for a response, it shows all the different assumptions that it made and the SQL that it used to generate it. Each data point gets an explanation — where it came from, what assumptions it made, and what SQL was used to get it. It might use the same SQL for all data points or different queries for different data points.
Q: Is the data on Quick Desktop somehow visible to the AWS team or the person’s team, or is it completely local? Is it used for training purposes or admin/control purposes?
A: Quick Desktop uses Bedrock, so things still go through CloudTrail and Bedrock — it’s not completely isolated. However, whether it’s web or desktop, we never use your questions or your data for training the models. That’s just not part of what we do. The knowledge graph is local at the moment, and the files that it accesses and builds are local as well.
Q: Can this handle data from multiple tools — ERP, Smartsheets, or multiple engineering tools — and suggest next steps on analytics?
A: Yes. If it has access to that tooling one way or the other — whether it’s through QuickSight data sets or QuickSight spaces, whether it’s through MCPs to other databases or other tools or other data catalogs — it’s able to scour those, leverage those, and use them to accomplish your goal.
Q: If any automation workflow is created, can that be audit friendly and authentic with respect to the inferences in performance to a hundred percent accuracy? (i.e., is it foolproof?)
A: A hundred percent accuracy and foolproof are tricky terms that probably require further discussion based on what you’re trying to do. We do have automations with log tracking for flows and automation — all of those get stored in the logs. For things that happen locally on Desktop, those get stored locally. How those are piped into CloudTrail logs will be clarified when the product goes GA. Please remember that Desktop is still in preview, with GA scheduled for shortly in the future.