# Agentic Architecture

Record: `arch:agentic-architecture`
Kind: pattern
Layer: [Correctness Core](https://banes-lab.com/records/layer/correctness-core.md)
Severity: contextual/mandatory in regulated domains
Scope: model-backed system, reasoning, decision flow
Canonical: https://banes-lab.com/ontology#arch-agentic-architecture

## Repair

- Refactored by: Bound the Agent with Tools, Limits, and Review
- Detected by: agent actions without tool scoping or step limits
- Violated by: an unbounded model loop acting with no guardrails
- Measured by: unguarded agent action rate
- Enforced by: model safety review

## requires

- [Model Inference](https://banes-lab.com/records/arch/model-inference.md)
- [Tool Interface](https://banes-lab.com/records/lex/tool-interface.md)

## reinforces

- [Explainability](https://banes-lab.com/records/arch/explainability.md)
- [Model Safety](https://banes-lab.com/records/arch/model-safety.md)

## enables

- [Bounded Tool-Using Agents](https://banes-lab.com/records/lex/bounded-tool-using-agents.md)
- [Governed Autonomy](https://banes-lab.com/records/lex/governed-autonomy.md)

## conflicts-with

- [Ungrounded Content](https://banes-lab.com/records/arch/ungrounded-content.md)

## tensions-with

- [Determinism](https://banes-lab.com/records/arch/determinism.md)

## tensions

- [Agentic Architecture Determinism](https://banes-lab.com/records/tension/agentic-architecture-determinism.md)