Glossary
🔍
77 terms · Page 1/8
Active Listening
Deliberate practice of observation and uncomfortable questions to reach the real problem beyond initial client articulation. The primary skill of Phase 1.
Active Listening goes beyond passive hearing. It requires asking questions the client may not want to answer, observing what is not said, and resisting the urge to propose solutions. In Phase 1, it is the primary mechanism for progressively discovering the Problem Statement. The practitioner must tolerate ambiguity and discomfort, knowing that premature clarity leads to The Solution Trap.
ADR
Architecture Decision Record
Architecture Decision Record. Formal document capturing a technical decision, its context, alternatives considered, and rationale for the chosen option.
ADRs are produced by the Architect Agent and reviewed by the Tech Lead. They provide an auditable trail of why technical choices were made, preventing revisitation of settled decisions. During Parallel Construction, Track Dev Leads reference ADRs to resolve ambiguity. ADRs feed into the Decision Log as persistent institutional knowledge.
Agent Chain
The specialized Agent pipeline: Analyst, PM, Architect, SM, QA. Each transforms previous output without contradicting it. The production line of Context Engineering.
The Agent Chain defines the sequence of transformations from Solution Brief to executable Story Files. Each Agent receives the output of the previous one and adds its specialized perspective: Analyst refines requirements, PM prioritizes, Architect designs structure, SM produces Story Files, QA validates. Downstream Fidelity is the invariant — no Agent may contradict, invent, or reinterpret what came before. Breaks in the chain signal context problems.
Agentic AI
AI systems capable of autonomous multi-step reasoning, decision-making, and tool use. The execution paradigm this methodology designs for from Phase 3 onward.
Problem-Driven AI embraces Agentic AI as the execution model but insists that agency without context is chaos. Agents are designed with specific Rules, Skills, and context boundaries. Agent Chains orchestrate multiple Agents in sequence. The methodology ensures agentic capabilities serve a validated problem, not operate autonomously without purpose.
Agents
Specialized AI profiles with defined roles, clear limits, and gap protocols. Not generic instances — each has identity, boundaries, and specific expertise.
Agents are the second dimension of Context Engineering. Each Agent in the chain (Analyst, PM, Architect, SM, QA) has a defined role, explicit boundaries, and a Gap Protocol. Agents transform the output of the previous Agent in the chain without contradicting it (Downstream Fidelity). They are not interchangeable — an Architect Agent cannot perform SM Agent tasks. This specialization enables quality control at each transformation step.
Analyst Agent
First agent in the chain. Refines raw requirements from the Solution Brief into structured, unambiguous specifications for downstream agents.
The Analyst Agent receives the Solution Brief and Context Document as input and produces refined requirement artifacts. Its Rules enforce completeness checks and terminology alignment with the glossary. Output quality here determines Downstream Fidelity for the entire chain, making it the critical first gate of Context Engineering.
Architect Agent
Third agent in the chain. Designs technical structure, produces the Dependency Graph, and generates ADRs for key decisions.
The Architect Agent transforms the PM Agent's prioritized plan into a technical blueprint. It generates the Dependency Graph, writes ADRs for non-trivial decisions, and defines component boundaries. The Tech Lead reviews its output. The resulting architecture directly shapes Story Files produced downstream by the SM Agent.
Aspirational Context
Anti-pattern: a Context Document that describes the ideal system rather than what has actually been built. The gap between documented context and reality grows silently.
Aspirational Context emerges when the Context Document reflects intentions rather than implemented reality. Features described as complete are still in progress. Market learnings are projected as applied changes before implementation. Agents operating on Aspirational Context make decisions based on capabilities that don't exist, producing output that cannot integrate with the real system.
Assumptions Register
Classified list of all assumptions: verified, low-risk, high-risk, unknown. Includes action plans for high-risk assumptions. Part of the Solution Brief.
The Assumptions Register forces the team to acknowledge what they believe but have not proven. Every solution carries assumptions — about market behavior, technical feasibility, data availability, user adoption. Classifying them by risk level and creating action plans for high-risk ones prevents the Solution Brief from being built on unstable ground. Unregistered assumptions become hidden risks that surface during Phase 4 construction.
BMAD
SacredBreakthrough Method for AI-Driven Development
Breakthrough Method for AI-Driven Development. Referenced framework that provides structural patterns for AI-driven development workflows.
BMAD is the referenced framework within Problem-Driven AI that informs the Agent-based architecture of Phase 3 onwards. Its patterns for specialized agents, structured skills, and governance rules are adapted and extended by this methodology. BMAD provides the execution machinery; Problem-Driven AI provides the problem-first philosophy that directs it.