Module AG10
Agents, tools, memory, and bounded execution
Build an agent only when its flexibility justifies added cost and failure risk.
Why this module
Build an agent only when its flexibility justifies added cost and failure risk.
What you produce
A tool-using assistant compared with a deterministic workflow. Assess the output against the unit outcomes, explanation, safe handling, and independent transfer. Do not require public sharing or use of real sensitive records.
Chapters and units
AG10-C01: Architecture and tool boundaries
Choose the simplest adequate execution design; Validate arguments and separate read from write capabilities
- AG10-U01 Workflows versus autonomous loops Choose the simplest adequate execution design
- AG10-U02 Tool schemas and authority Validate arguments and separate read from write capabilities
AG10-C02: State and durable control
Define what is stored, for how long, and with whose authority; Resume safely and stop repeated or over-budget actions
- AG10-U03 State, context, and memory Define what is stored, for how long, and with whose authority
- AG10-U04 Checkpoints, stop conditions, and approval Resume safely and stop repeated or over-budget actions
AG10-C03: Protocols and failure analysis
Explain protocol boundaries and a justified handoff; Test malformed calls, untrusted tool output, and duplicate actions
- AG10-U05 Protocols and orchestration patterns Explain protocol boundaries and a justified handoff
- AG10-U06 Agent failure and comparison lab Test malformed calls, untrusted tool output, and duplicate actions