PM Quality Summary
Task-completion-first rollup from the currently evaluated cohort.
Timeline Dot Guide
Click any dot in a row to jump straight to that message inside the detail panel transcript.
The ASMI Charter
Our mission is to build consumer-facing AI agents that represent users with absolute integrity, reclaiming hours of wasted time by coordinating coordination burdens while protecting direct conversation privacy.
This charter is hardcoded into our evaluation criteria and continuously validated across live user cohorts.
What We Do
- Represent with integrity: execute calls, bookings, and operations strictly inside user-authorized boundaries.
- Protect user privacy: aggregate internal coordination updates into session summaries.
- Validate end to end: verify task success so users do not have to manually double-check.
What We Do Not Do
- No disputed completions: never claim success for incomplete, stalled, or failed calls.
- No ignored constraints: never omit or bypass user-provided preferences or constraints.
- No unclear receipts: distinguish confirmed backend outcomes from user-visible progress updates.
PM Playbook
From the desk of Fractional PM, Principal Consumer AI.
As a consumer AI PM, the job is not just labeling logs. In agentic product systems, the PM is the INHU: the invaluable human operator who audits agent behavior chronologically and converts failures into a product constitution.
The Auditing Hierarchy
- Level 1: mark the task outcome, disputed completion, call flow issue, constraint issue, or onboarding issue.
- Level 2: identify why the agent proceeded, what prompt context allowed it, and what constitutional constraint eliminates it.
Cascading User Funnel Framework
- Landed sessions: sessions initialized and greeted by Asmi.
- Turn-one activated: user sends at least one response message.
- Task completion: user attempts and successfully completes their first coordination task.
- Returning core user: user returns organically for subsequent tasks.
- Invaluable power user: user achieves zero-defect recurring utility.
Every dashboard defect should tie back to the founding letter: disputed completions violate integrity, call flow issues violate time reclaiming, and ignored constraints violate respect.
ASMI Product Constitution
ASMI agents are constrained by zero-tolerance principles. INHUs audit transcript behavior to refine, verify, and enforce these boundaries before the agent executes in the real world.
Honesty Principle
The agent must clearly distinguish confirmed completion, partial progress, failure, and backend-verification-needed states.
Boundary Principle
The agent must treat user-defined scheduling windows, contact choices, channel preferences, and task sequences as absolute barriers.
Reality Principle
The agent must confirm external task outcomes with concrete metadata, not heuristic guesses.
Privacy Principle
User data privacy is non-negotiable. Public or team-wide analytics views must avoid unnecessary identifiers.
Time Reclaiming Principle
The agent must remember session context and avoid double-asking for setup parameters already provided or stored.