Context

Friction analysis of 148 documented friction points revealed two unsolved paradigm-level problems:

1. Person-centricity (hierarchy vs graph mental model) - causes multi-call workflows
2. Stale facts (point-in-time truths in call notes become outdated) - causes wrong answers

Both cause agents to give wrong answers. But stale facts is more dangerous:

• Invisible: Agent doesn't know the info is outdated
• Trust-eroding: User corrects agent, questions reliability
• Demo-killing: Wrong answer during live demo = failed opportunity

The Claire Vo Factor

Follow-up deadline: end of January 2026. She hosts live demos on "How I AI" podcast.

The Wessel start date error is the canonical example:

• Agent reported "Jan 15" (from Dec 16 call notes)
• Actual answer was "February" (from Dec 18 decision)
• The decision existed but didn't formally supersede the call note
• No canonical source of truth for entity facts

If this happens during a Claire Vo demo, the opportunity is lost.

Decision

Prioritize resolving stale facts over:

• New compound tools (prepare_for_meeting can wait)
• New features (semantic search, embeddings)
• Cleanup work (archiving isolated tasks)

Specifically: Progress "Person scopes as canonical facts" exploration (7e3e346) to a decision and implementation.

Implications

1. Person scopes become canonical source for entity facts
2. Agents check Person scope first for questions about people
3. Call notes remain raw capture; Person scope holds processed truth
4. When facts change, Person scope is updated

What This Deprioritizes

• prepare_for_meeting compound tool - wait for evidence primitives are insufficient
• Semantic/fuzzy search - too much complexity for current state
• Additional MCP tools - 58 is enough, refine before adding
• Isolated filter / bulk archive - nice cleanup but doesn't affect demo quality

Context

Friction analysis of 148 documented friction points revealed two unsolved paradigm-level problems:

1. Person-centricity (hierarchy vs graph mental model) - causes multi-call workflows
2. Stale facts (point-in-time truths in call notes become outdated) - causes wrong answers

Both cause agents to give wrong answers. But stale facts is more dangerous:

• Invisible: Agent doesn't know the info is outdated
• Trust-eroding: User corrects agent, questions reliability
• Demo-killing: Wrong answer during live demo = failed opportunity

The Claire Vo Factor

Follow-up deadline: end of January 2026. She hosts live demos on "How I AI" podcast.

The Wessel start date error is the canonical example:

• Agent reported "Jan 15" (from Dec 16 call notes)
• Actual answer was "February" (from Dec 18 decision)
• The decision existed but didn't formally supersede the call note
• No canonical source of truth for entity facts

If this happens during a Claire Vo demo, the opportunity is lost.

Decision

Prioritize resolving stale facts over:

• New compound tools (prepare_for_meeting can wait)
• New features (semantic search, embeddings)
• Cleanup work (archiving isolated tasks)

Specifically: Progress "Person scopes as canonical facts" exploration (7e3e346) to a decision and implementation.

Implications

1. Person scopes become canonical source for entity facts
2. Agents check Person scope first for questions about people
3. Call notes remain raw capture; Person scope holds processed truth
4. When facts change, Person scope is updated

What This Deprioritizes

• prepare_for_meeting compound tool - wait for evidence primitives are insufficient
• Semantic/fuzzy search - too much complexity for current state
• Additional MCP tools - 58 is enough, refine before adding
• Isolated filter / bulk archive - nice cleanup but doesn't affect demo quality