Chatting with a model about your data is one thing. You paste a table, ask a question, get a paragraph. Agentic analytics is a system built around the model. It connects to the warehouse and reads the metric definitions you wrote. It plans the analysis, runs the SQL, makes the chart, and drafts the memo. Then it publishes where your team already reads. It also checks itself along the way, if you built the checks.
What you get depends on the system around the model more than on the model. The same model gives a different quality of answer depending on what sits around it. The skills that encode how your team analyzes. The context store with your definitions. The tools it can call. The eval suite that tells you when it is wrong. Swap the model and most of that stays. That is why the course spends one week building the analyst and four weeks on everything around it.
A concrete example. A VP writes "checkout conversion dropped, can you look into it?" A chat model asks what data you have. An agentic analyst pulls the funnel from the warehouse and decomposes the drop by segment and step. It checks the segment sizes, produces the chart, and writes a memo with the query behind each number. You read the memo and decide. You still make the call. The pulling and checking moved to the system.
The failure modes are different too. An agentic system rarely crashes. It returns a confident number computed on the wrong definition, or on the wrong date field. Or on a sample that was filtered without telling you. Shane's post on ten ways an AI analyst breaks lists the ones we keep hitting in real companies.
In the courses
Agentic Analytics: Build an AI Analyst is five weeks on this. Week 1 (Foundations of agentic analytics) builds the first skill and a small agent system. Week 2 (Design the system, connect MCPs and a warehouse) connects it to data and to your team's tools. Weeks 3 to 5 measure it, give it context, and swap the engine. AI Analytics for Everyone is the other side. You use these systems rather than build them, and you check the answers yourself.