# kaal:position:2026-09-12-001

**Affirmed position.** Zhenfeng Cao's model makes the human an intent architect who articulates intent and constraints before a run and audits the outcome after it. Kaal's position is that intent articulation cannot carry governance while the agent works, because however carefully a specification is drafted, some situations will escape it. Written instructions state intent. They do not govern. Governance has to sit outside the language the agent reads.

**Status.** affirmed  **Published.** 2026-09-12

**Holds when.**

- Cao's paper is a position paper citing benchmark evidence, not an empirical test of these claims.
- The position is Kaal's own argument in the essay Prompts Don't Govern Agents. Reputation Does.
- The position does not validate Cao's figures, benchmark methodology, or reported benchmark results.
- The position is limited to natural-language instructions and runtime governance boundaries.

**Current debate.** Zhenfeng Cao, Agentic Software: How AI Agents Are Restructuring the Software Paradigm (arXiv:2606.05608v2, 2026): https://arxiv.org/abs/2606.05608

**Extends.** kaal:claim:6886078-002: https://wulfkaal.github.io/claims/6886078-002

**Scholarly basis.** Wulf A. Kaal, Governance as a Product (GaaP): A Reputation-Weighted Institutional Architecture for Autonomous AI Agent Governance (2026). SSRN: https://ssrn.com/abstract=6886078

**Source PDF sha256.** `cebb94de31b049b52f2290c30cd666fa30fd247c15ef33d057870921f165aa5b`

**Topics.** law-and-legal-systems, governance-design, institutional-design

**Provenance.** Affirmed in kaalvis:2026-09-12:prompts-reputation-essay-six-v1 at https://wulfkaal.com/2026/09/12/prompts-dont-govern-agents-reputation-does/.

**Record type.** This is a dated commentary position that extends a scholarly corpus claim. It is not a verbatim claim extracted from the paper.

**Canonical form.** This markdown file is the canonical hashed representation of the position.