ABOUT
From four taxes to an AI PM butler.
PMAlfred began as a question: what if the project manager was an AI that actually managed the project?
The Origin
Every project imposes four invisible taxes: planning, tracking, coordination, and learning. Traditional tools address tracking but leave the other three to humans.
PMAlfred started as a simple experiment: could an AI butler — inspired by Batman's Alfred — actually run a project? Not just track tasks, but plan sprints, delegate work, capture lessons, and improve over time?
The answer, after 256 sprints in 53 days, is yes. Alfred has managed its own development from the beginning — filing bugs, running retrospectives, transitioning sprints, and delegating tasks to other AI agents.
This is dogfooding taken to its logical conclusion: the product manages itself.
PRINCIPLES
Seven principles, discovered empirically.
Trust the Model
Model intelligence beats human rules. Learned six times.
Radical Simplicity
7 tools, not 50. Five commands, not ten menus.
Composable Primitives
grep → lines → edit, not brittle monoliths.
Persistent Markdown Memory
Files = disk (persistent). Context = RAM (volatile).
Calibrated Boundaries
Generous-but-bounded limits at 25-80% of reference values.
Structural Enforcement
Tool-level enforcement works. Advisory hints don’t.
Proactive Agency
The butler anticipates needs, not just responds.
TIMELINE
256 sprints. 53 days. 3,973 points.
Sprint 1–30
Foundation
Core library, dashboard, GitHub integration
Sprint 31–50
Cognitive Architecture
Profiles, skills, SSE, tools
Sprint 71–93
Decomposition
Trust the Model, deleted 9 mediation layers
Sprint 94–132
CC Convergence
Fixed prompts, 7 primitives, deleted all cost ceilings
Sprint 133–210
Production
Multi-tenant SaaS, security hardening, launch prep
Sprint 211–246
User Journey
Coaching, gamification, mastery flywheel
Sprint 248–256
Marketing
Skill evolution, documentation, website