Sharing AI knowledge, practical implementation records, and more.
The AC/DC framework reimagines software development by moving verification from post‑commit CI pipelines to the moment code is generated, letting AI agents write, check, and fix code in a self‑correcting loop, while redefining human engineer responsibilities.
The article explains how Spec‑Driven Development (SDD) transforms AI‑generated code from risky demos into production‑ready features by defining executable specifications, enforcing review, injecting context, and automating verification, illustrated with a concrete order‑export example.
LLM Wiki, proposed by Andrej Karpathy, replaces repetitive RAG retrieval for personal agents with a three‑layer markdown‑based knowledge base that separates raw sources, curated wiki pages, and schema constraints, enabling durable, auditable memory, structured updates, health checks, and a hybrid Wiki‑RAG workflow.
Claude Code tackles the bottleneck of overloaded main sessions in complex code tasks by splitting work into a Coordinator that keeps the overall goal and independent Workers that handle research, implementation, testing, and review in isolated contexts, returning only essential evidence for synthesis.
The article explains why feeding vague PRDs to AI coding tools leads to unstable code, and presents a step‑by‑step workflow that rewrites rough requirements into a structured AI‑First PRD covering goals, boundaries, states, acceptance criteria, and feasibility checks before implementation.
The article explains why teams using multiple AI coding agents need a top‑level CONSTITUTION.md to capture stable engineering principles, decision hierarchy, autonomy boundaries, quality gates, and revision processes, and shows how to structure, write, and propagate it across tool‑specific files.
The article breaks down Claude Code’s persistent agentic loop into four layers—QueryEngine, Tool System, Permission/Hook, and React + Ink—explaining how each turn gathers context, makes model decisions, executes actions, verifies results, handles errors, and renders a terminal UI.
The article explains how Hermes structures the runtime execution of agents—capturing tool calls, context changes, results, and rewards—so that these trajectories can be evaluated, fine‑tuned, and fed into reinforcement‑learning loops, creating a continuous improvement cycle.
The article dissects Hermes’s extension layer, showing how the Tool Registry converts functions into runtime assets, how AST‑based auto‑discovery and availability checks keep tool lists accurate, and how the unified dispatch, Plugin System, Hook System, and replaceable Context Engine together form a robust, governable agent platform.
Hermes Prompt Runtime introduces a layered architecture that first resolves the model provider, then builds a stable system prompt, freezes memory snapshots for session boundaries, isolates per‑call temporary context, and compresses long histories, thereby keeping long‑term semantics stable, improving prompt caching, and reducing context‑window pressure.
