Agentic Capabilities
What you must understand
Agentic capabilities extend automation with agents that plan, call tools, and act on results. Know the loop: observe, decide, act, and reflect. Understand tool definitions, function calling, and how an agent selects a tool. Be fluent in guardrails, human-in-the-loop approval, and audit logging. Know how agents differ from deterministic processes: agents choose steps at runtime, so you need bounds, allowlists, and termination conditions. Also understand memory and context management, and how to hand off between an agent and a standard process.
Traps that appear
Questions often test when to use an agent versus a deterministic process. Another trap is assuming an agent will always terminate; without a step limit it may loop. Watch for questions about tool permissions and whether an agent can call any connector or only allowlisted tools. Also expect traps around human approval placement and audit trail completeness.
Worked example
You need an agent to triage requests and open tickets. The rule of thumb: allowlist tools, cap steps, require approval for writes. Define read-only tools for lookup, a write tool for ticket creation gated by human approval, and a maximum step count. Log every tool call with inputs and outputs so the decision path is auditable.
Rule of thumb
Give agents the smallest tool set and the tightest bounds that still solve the task.