Describe what an MCP-connected AI assistant can and cannot reliably do for business configuration and analysis tasks
Recognize when a task exceeds what this track covers and should be escalated to a developer or architect
This Knowledge Track is about building the judgment you'll need to effectively make use of AI assistants and related tooling to speed up processes without needing to be technical. Before beginning, it is important to remember that every AI output you produce in this track is a starting point for your review, never a finished deliverable. AI tooling can produce highly polished content that looks correct at a glance, but is actually inaccurate.
How and When to Escalate
If you've tried refining your request two or three times and the output still doesn't look right, that's usually a signal the task has moved outside what you can safely resolve on your own, so it's usually a good idea to stop and ask someone who is more technical to assist.Escalating well is also a skill in itself, and it's worth setting the bar early rather than leaving it vague. A good escalation to a developer or architect typically includes:
What you were trying to accomplish (the original business question or request)
What you tried, including the specific prompts or requests you used
What came back, and specifically what didn't look right or couldn't be resolved
What you've already ruled out, if anything
Compare "this dashboard thing isn't working" against "I asked for a widget showing weekly return volume by reason code; the draft generated correctly but the data source it referenced returned an empty result even after I re-specified the date range twice. Here's exactly what I asked and what came back." The second version respects the receiving developer's time and gets you a faster, better answer.
Treat AI Generated content as a draft
Remember to manually review all results returned by an LLM. Mistakes may not always be obvious at first glance.