What should an AI agent do next?
Investigate a surprising metric. Roll back a failing model. Decide when a human should step in. Practice the decisions data scientists make when agents enter the workflow.
Choose a learning path ↓
Choose your path
Agentic analytics
Average ticket spend jumped from $840 to $1,404. Trace the agent’s explanation, then test whether the evidence supports it.
Trace the agent’s answer
Pick a question and inspect six components, from the metric definition to the briefing.
Open the map →Catch a convincing mistake
Run four test cases. Inspect a failure, change the agent’s policy, and see which answers improve.
Run the evaluations →Try a second dataset · Evaluate a retail search query →
Machine learning in production
A notebook gives you a model. What happens when an agent can retrain it, send it traffic, or flag a bad forecast? Follow a retail demand model through three production decisions. Set the evidence each agent needs and decide where an engineer steps in.
Could an agent reproduce your training run?
Build a run manifest an agent can replay and an engineer can audit.
Open lab →Should the agent roll out this model?
Route a canary and decide when the deployment agent should roll back.
Open lab →When should a monitoring agent intervene?
Separate a change in traffic mix from a regression in forecast quality.
Open lab →
Applied ML in an Agentic World
Try three decisions that come up in model development and review.
When does 80% actually mean 80%?
Keep the shoppers and their purchases fixed. Change the confidence of the model that scored them.
Open lab →Which decisions should a person review?
Give a reviewer 30 slots and a batch of 100 product labels. Decide which labels reach their desk.
Open lab →Would you buy what your search returned?
Run one grocery query three ways. Inspect which products satisfy the shopper’s whole request.
Open lab →