Tools, planning, state, memory, and autonomous actions

Overview

• Tools are approved external capabilities.
• Planning breaks a goal into steps; state records current execution facts; memory carries selected information across turns or tasks.
• Autonomous action means the system can perform some actions without a new user choice at every step, subject to its assigned policy.

Why this matters

Multi-step tasks need progress tracking and enough continuity to avoid asking the same questions repeatedly.

An everyday example

An employee uses a task list, a case file, and a limited set of permissions.

A technical example

State includes claimId, missingFields, draftId, and phase; memory stores an approved communication-language preference.

How it works

1. Initialize trusted state
2. Update after verified events
3. Create/adjust bounded plan
4. Retrieve relevant memory
5. Execute permitted steps
6. Checkpoint
7. Validate completion.

Architecture and flow

Task state + selected memory → planner → tool executor → verified state update.

The insurance assistant example

Read-only lookups may proceed automatically; submitting a claim requires explicit user confirmation of the exact draft.

When to use it

Multi-turn assistance and adaptive information gathering.

When to avoid it

Do not persist guessed facts as verified memory or give a model general-purpose credentials.

Interview answer

“State tracks this run, memory provides selected continuity, and planning coordinates allowed steps. Autonomy is a scoped permission policy.”

Benefits, trade-offs, and production considerations

• Plans become stale after tool errors or user corrections.
• Treat plans as proposals, bind actual writes to current validated state, and keep memory relevance and retention explicit.