Research Note

Agent Action Authorization Research Note

Meeting-derived action should move through explicit states: candidate, confirmed commitment, assigned owner, approved write, executed change, verified result, and recover

Aug 4, 20261 min readBy Dalton Anderson

Source conversation

QUILL: AI CHIEF OF STAFF WITH MICHAEL DAUGHERTY

keywords AI agents, Chief of AI Staff, privacy, localized models, MCP, automated workflows, meeting management summaryQuill defines a Chief of AI Staff as a tool that manages work processes and AI agents rather than managing people. AI aims to reduce coordination overhead toward zero, enabling smaller, high-judgment teams to execute at scale. Privacy is a core differentiator, with Quill utilizing local models and storage to ensure users maintain absolute control over their sensitive data. Personalization is achieved through conversation; by analyzing meetings, Quill learns user preferences and tool usage to automate follow-ups and task creation. The Model Context Protocol (MCP) allows Quill to explore tools like Airtable or Obsidian to build its own internal instructions. Closing the loop with AI allows for rapid iteration, such as running tiny research experiments overnight to improve performance without constant human intervention. AI can serve as a proactive teacher, with Quill using internal agents to onboard team members and teach codebases through personalized curricula. Strategic human judgment remains the essential North Star, especially in creative or high-stakes domains where AI-generated content requires oversight. Host Dalton Anderson interviews Michael Daugherty, CEO of Quill, regarding the Chief of AI Staff vision. Quill transforms messy meetings into actionable workflows while prioritizing privacy through localized data storage. Daugherty explains how Quill uses the Model Context Protocol (MCP) to integrate with production tools like Linear and Notion, learning user intent directly through natural conversation. They discuss closing the loop to allow AI to self-iterate and the use of AI agents to reduce tribal knowledge by teaching codebases to new team members. The episode highlights a future where AI handles execution, leaving humans to focus on high-level strategy and judgment. sound bits "A chief of AI staff is ultimately going to manage the way we work in the future." "All that coordination overhead can go down very, very close to zero eventually because AI can actually do this execution." "As soon as you complete the loop with AI, it can iterate on itself." "Software can teach you how to use itself and make everybody an expert user."

Back to the episode hub

Agent Action Authorization Research Note

Meeting-derived action should move through explicit states: candidate, confirmed commitment, assigned owner, approved write, executed change, verified result, and recoverable record.

A transcript can be wrong. A speaker can brainstorm without committing. A participant can commit without authority to bind another person. A correct commitment can still belong in the wrong system.

NIST's AI RMF Measure playbook recommends auditability, histories, oversight measures, override records, and documented go or no-go decisions. NIST's 2026 agent identity and authorization concept paper frames agent identity, delegation, least privilege, auditing, and human-in-the-loop authorization as open enterprise design problems.

The smallest safe write uses a named human or service identity, minimum scope, a complete preview, idempotency or duplicate protection, a returned record identifier, and a reversal or correction path.

Human review must occur at the authority transition, not merely somewhere in the workflow.

Sources

Follow the evidence.

security guidancemodelcontextprotocol.io
18 U.S.C. 2511law.cornell.edu
agent identity and authorization concept papernccoe.nist.gov
local-first software essayinkandswitch.com
device encryption guidancecisa.gov
meeting recap studymicrosoft.com
MCP specificationmodelcontextprotocol.io
reader-focused meeting summarization researchaclanthology.org
authorization guidemodelcontextprotocol.io
agent evaluation worknist.gov
current Quill documentationquillmeetings.com
key-management guidancecsrc.nist.gov
server conceptsmodelcontextprotocol.io
California Penal Code section 632leginfo.legislature.ca.gov
AI RMF Measure playbookairc.nist.gov
data sovereignty pagequillmeetings.com
current About pagequillmeetings.com
OWASP's MCP security guidancecheatsheetseries.owasp.org
SP 800-209nist.gov
query-focused meeting summarization researchaclanthology.org