Build AI that works with business context.
Builds governed retrieval, agents, tools, and automations with evaluation, permissions, and human approval boundaries.
The context behind the work.
You combine prompts, retrieved knowledge, tools, and application behavior into a useful AI feature. The surrounding context determines whether the output is dependable: source quality, access boundaries, evaluations, and human review all belong in the design.
Build an assistant over business knowledge
Select approved sources and define the task and audience. Test whether answers are grounded in the retrieved material, check access boundaries, and make the expected behavior clear when relevant evidence is missing.
Add a tool that can change a business record
Define the allowed action, required permissions, and human approval point. Evaluate ambiguous and failure cases before release, then retain the prompt and tool configuration used by the feature.
A practical path from task to outcome.
Deliver an AI feature with versioned prompts, evaluation evidence, and controlled access.
- 01
Design the approach
Define the user task and identify the approved knowledge sources and tool permissions.
- 02
Build the workload
Build the prompt or agent workflow and keep prompt versions and input expectations explicit.
- 03
Evaluate quality and risk
Evaluate representative cases, including failure and boundary conditions, and compare quality with cost.
- 04
Automate controlled action
Release the reviewed version through the agreed approval path and observe behavior after integration.
A versioned AI workflow with evaluations, permissions, and a named owner.
Less repeated effort. More useful work.
Explore the habits and platform connections that can make this role easier, more consistent, and easier to collaborate with.
Prompt logic scattered through application code
Reference versioned prompts and keep the evaluation context with the release.
Repeating evaluation setup for every change
Reuse a reviewed evaluation set and compare candidate versions consistently.
Unclear responsibility for AI actions
Define tool permissions and the human review points before release.
Measure your own improvement
Choose a baseline before you begin. Review these signals with your team; results depend on your data, process, and implementation.
- Time from a prompt change to a reviewed evaluation
- Escalations caused by missing context or permissions
Build confidence with a first task.
Deliver an AI feature with versioned prompts, evaluation evidence, and controlled access.
Use AI with judgment
Use AI to draft candidate prompts and test cases; review retrieved evidence and keep consequential actions under human control.
Your practice checklist
0 / 4 completeThe right surfaces. The right people.
Continue into the product, deepen your knowledge, or follow the next role in the handoff.
Explore the product surfaces
Cortex AI & AgentsGeneCatalog & KnowledgeData Science & MLOpsGovernance & ApprovalsGo deeper
Technical workbookDocumentationTechnical workbooks are maintained in English. Workspace access and available capabilities depend on your deployment and permissions.
Bring your own workflow.
Explore how these practices could fit your team, your data, and your operating requirements.