1
Define the goal and constraints
Write a one‑sentence outcome, guardrails, and what success looks like.
2
Provide context and memory
Store domain facts, glossary, and examples. Persist short‑term state between steps.
3
Equip tools (MCP)
Expose the data sources and actions the agent can use (e.g., search, tickets:list, analytics:topIssues). See
/learn/mcp.4
Implement the reasoning loop
Plan → call tools → observe → revise plan until done. Keep outputs structured.
5
Deliver results
Return a clear artifact (report, JSON, message) and log decisions for auditability.
When your agent reliably achieves the goal with safe defaults and clear logs, you’re ready to automate.
