Engineering Trustworthy Agentic AI Systems
Governed Autonomy, Intelligent Supervision, and the Protection of Business Trust
Agents may work. Agents may supervise. Neither may govern itself.
Executive Perspective¶
Agentic AI creates extraordinary opportunity—and the possibility that a plausible model inference becomes a consequential business action at machine speed. This paper develops an architecture of governed autonomy: worker and supervisory agents operate within bounded authority; independent trusted services enforce policy and protect authoritative state; and evidence, containment, reconciliation, and recovery keep business outcomes accountable.
What You Will Take Away¶
- Distinguish an agent assertion from an authorized, verified business outcome.
- Explain the two-gate execution boundary and why supervision is not security authority.
- Design authoritative state, evidence provenance, and bounded interventions.
- War-game excessive refunds, false completion, malicious telemetry, retries, and common-mode failures.
- Balance business value, control cost, assurance, and recoverability.
Who Should Read This Paper?¶
Engineering leaders · Enterprise architects · Technology executives · Security and governance teams · Advanced students.
Read the Companion Paper¶
COMP-WP-006 — Becoming a Software Engineer in the AI Era translates the architecture and industry obligations into capabilities students can develop and demonstrate.
The two papers form a reading pair: what organizations must engineer and what emerging engineers must become. For students, read the technical paper first to see the real obligations, then use the education paper to plan and document professional growth.
Suggested Further Reading¶
Citation¶
IEEE
W. T. O’Connell, “Engineering Trustworthy Agentic AI Systems: Governed Autonomy, Intelligent Supervision, and the Protection of Business Trust,” ETIS White Paper Series, WP-013, ver. 1.0, Oct. 2026.
APA
O’Connell, W. T. (2026). Engineering Trustworthy Agentic AI Systems: Governed Autonomy, Intelligent Supervision, and the Protection of Business Trust (WP-013, Version 1.0). Engineering Trustworthy Intelligent Systems. https://etisframework.org/publications/white-papers/wp-013/
Canonical URL: https://etisframework.org/publications/white-papers/wp-013/
Version History¶
| Version | Date | Description |
|---|---|---|
| 1.0 | October 2026 | Initial publication; ETIS companion-paper edition |
The PDF is the authoritative public edition. The Word file is retained as an editable publication master.