PAAP Framework
Designing an AI agent framework that replaces the process overhead killing product teams — so PMs can focus on decisions, not admin.
The System
PAAP (Product-Aligned Agent Process) is 10 specialized AI agents covering 4 lifecycle phases. Each agent has an identity, behavioral constraints, and rules it cannot break.
10 Specialized Agents
Product Strategy
Challenges ideas, runs discovery
Hypotheses
Validates or kills hypotheses
Master
Orchestrates all agents
Architecture
Code quality gate
Backend
APIs, DB, business logic
Frontend
UI components and pages
Design
UX flows and specs
Metrics
Tracks outcomes, feeds learnings
Marketing
Copy, social, newsletters
Prompt Engineer
Crafts and reviews prompts
The whole system runs in a terminal. Each agent is a markdown file. No custom models, no fine-tuning, no proprietary infrastructure. Just very specific instructions given to a general-purpose model.
Beyond agents, the framework includes 7 slash commands for common workflows (hypothesis-to-PRD, sprint reports, velocity analysis), 10+ templates standardizing artifacts across Jira and Confluence, a Figma plugin for visual process documentation, and an automated installer with a full onboarding guide.
Agents in Detail
Product Strategy Agent
Challenges product ideas. Its job is to pressure-test, not validate.
When it evaluates an idea, it runs through Cagan's four product risks (value, usability, feasibility, viability), maps the opportunity space using Torres's Opportunity Solution Tree before jumping to solutions, and applies Christensen's Jobs to Be Done to check for real struggling moments.
Sprint Report Automation
Turned 90 minutes of manual assembly into 40 seconds.
The Master Agent triggers the report workflow: pulls completed stories from Jira, matches them to feature-level metrics in PostHog, compares against PRD targets, and flags anything off track. The report doesn't just summarize — it recommends: investigate, keep monitoring, or escalate.
Real Deployment
PAAP isn't a concept — it's deployed. We adapted it for CreatorIQ's design team as PAAP-CreatorIQ, with custom agents tailored to their workflows.
Custom Agents
Design Feasibility Check, Design-to-Eng Handoff, Systems Architecture Advisor — replacing generic agents with role-specific ones.
Team Workflow
Daily, weekly, and monthly rituals integrated into the design team's existing process — not replacing how they work, but augmenting it.
PAAP is now the core of Floe Studio — our productized offering of Agentic Discovery & Delivery as a Service.
Context
Roughly a third of every PM's week went to process administration. Writing Jira tickets, copy-pasting acceptance criteria, pulling PostHog numbers into spreadsheets, formatting reports that three people skim and nobody acts on.
The underlying problem is that PMs are doing work a system should handle. Sprint reports, PRDs that inherit hypothesis context, backlog items generated from validated opportunities — none of this requires judgment. All of it requires time.
Key insight
Constraints beat capabilities. A general-purpose AI with very specific instructions outperforms a specialized tool with loose guidance.
Want to discuss systems design + AI?
I'd love to talk about how design thinking applies to AI agent systems.
Book a call