A 311 resident-services desk designs an assistant that answers permit status from approved knowledge and may call tailored actions. Which option fits?

A 311 resident-services desk that must answer permit status from approved knowledge and may call tailored actions should use a custom agent grounded on that approved knowledge—extending beyond built-in Copilot alone when status lookups and actions require curated sources and tools. Custom agents let municipalities bind answers to known permit knowledge and invoke tailored actions that stock Copilot experiences do not provide out of the box.

Built-in Microsoft 365 Copilot grounded on approved knowledge for permit-status answers with standard chat summarization fails because standard chat summarization cannot execute the tailored permit-status actions the scenario requires. A Copilot Studio agent grounded on approved knowledge for permit-status answers with prebuilt shared connectors and prompts fails because prebuilt shared connectors and prompts do not provide the tailored actions required for permit-status lookups. A custom agent grounded on approved knowledge for permit-status briefs with built-in chat summarization and history context falls short because it relies on built-in chat summarization and history context instead of the dedicated tailored actions the scenario demands.

Exam caveat: AB-900 introduces custom agents at a conceptual level—grounding, tools/actions, and when built-in Copilot is insufficient—not full agent-studio implementation detail. Agents still inherit identity, DLP, and permission boundaries.

Operational check: Prototype a 311 permit-status agent against a curated knowledge source and a test action (status lookup), compare answers to built-in Microsoft 365 Copilot with standard chat summarization, and confirm prebuilt shared connectors and prompts are not substituted for tailored actions.