Understanding and Reviewing AI Work
AI can make mistakes. Before you rely on an answer or a change, check it the way you'd check a teammate's work.
Check what it used
- Open the linked objects. Systems, Resources, artifacts, and jobs in a reply appear as chips, small linked labels you can open. A struck-through chip couldn't be opened: it may not exist, you may not have access to it, or the platform couldn't check.
- Review the sources. Select Show artifacts to list the Resources the conversation referenced or produced.
- Confirm the context. Make sure the answer is about the right System, branch, and revision.
Check what it changed
- Revisions: compare the revisions it refers to on the Resource's Compare tab.
- Lineage: open the Lineage tab to trace which model and job produced the artifacts behind the answer.
- Jobs and workflow logs: check the status and outputs of the work the answer depends on.
- New files: review a drafted file, including its control tags and, when InfoSec levels are on, its InfoSec level, before you save it.
Know its limits
- AI Chat only sees what you can access. Control tags and InfoSec levels still apply, so an answer may leave out Resources you can't open.
- An answer is a starting point, not evidence. The evidence is in the Resources, revisions, and lineage it points to.
In the demo, the engineer checks the AI's answer in Lineage. They find the Rev1 baseline, the Rev2 revision, and the artifacts extracted from each, such as named_cells.json, to confirm the values the AI compared.