ADPS Knowledge Base

Topics

Patterns address recurring local structures. Topics organize engineering questions that cross several modules.

Each topic states its terms, practice sources, links to patterns and cases, and unresolved questions. See Workshops for the underlying discussions.

Current topics

The topic programme currently covers three areas: AI-driven software engineering, observability practice, and agent evals and testing. These discussion drafts will be revised with further cases and expert practice.

Topic research draft

AI-Driven Software Engineering: Specifications, Context, Execution, and Feedback

A shared engineering frame for AI coding, SDD, and agentic software engineering across specifications, context, execution, verification, and evolution.

AI software engineeringSDDHarnessCross-cutting

Topic research draft

Observability-Driven Agent Evolution: From Runtime Timelines to Verifiable Changes

Runtime trajectories, component versions, external outcomes, and change evidence for debugging, release, and agent evolution.

ObservabilityAgent evolutionTraceCross-cutting

Topic research draft

Agent Evals and Testing: From Output Scores to System Acceptance

Deterministic tests, trajectory evaluation, external acceptance, and production feedback combined into agent release evidence.

Agent evalsTestingAcceptanceRegression

Research agenda

Enterprise Agent Evolution and Operating Model: Research Agenda

A research agenda for the concurrent evolution of agent authority, engineering platforms, and organizational responsibility.

Enterprise evolutionAutonomyOperating modelResearch agenda

Topic lines in development

Long-running agent engineering: Goal retention, progress evidence, checkpoints, memory recall, budgets, and recovery.
Agent asset lifecycle: Specifications, skills, memory, evaluations, certification, and invalidation.
Event-driven execution and choreography: Event ingress, execution topology, idempotency, compensation, and the boundary of decentralized coordination.

These pages will be published after cases or workshops provide enough material.