Langmerge Agentic Systems Lab — The Playbook

Spec-Driven Development

Spec-driven development means writing the spec first — the constraints, the canonical definition, the runbook — and holding the AI to it instead of letting it improvise. This is the practitioner's guide: how we actually do it, start to finish.

killdate-kit is the repo we fork every time we start something new — a pre-wired CLAUDE.md, Cursor rules, and a runbook system, so the AI tools have structure to push against. These posts walk through how we use it. Best read in order.

Fork killdate-kit on GitHub →

Foundations

00 Agentic orchestration: what it is, what it isn't orchestration 01 Technical Knowledge Floor tooling 02 killdate-kit: a tour setup 03 The broader toolset tooling

The Minimum Viable Build

05 CLAUDE.md / AGENTS.md: pre-build session discipline claude-md 06 Define the product. Constrain the AI. product-definition 07 Scoping the minimum viable build (MVB) in Claude Code mvb 08 Shipping the Minimum Viable Build runbook

Feature Development

13 Agile development in an AI environment agile 14 Scoping the feature: lock, blast radius, mocks pre-build-lock 15 Building a feature: runbook, Cursor, Strangler Fig runbooks 16 Shipping a feature: staging, flags, kill dates shipping

Testing

17 Testing as part of the build testing 18 When to test, when not to, and knowing your app testing

Prompts in Production

25 Treat your prompts like code prompts 26 When your LLM output stops being what you shipped drift

Tooling: MCP

08 What Is the Model Context Protocol (MCP)? mcp 09 MCP servers for your build stack tooling 10 Building an MCP Server in One Autonomous Session (Case Study) mcp 11 How to Build an MCP Server (Custom, in Python) mcp

Designing with AI

19 Figma + Claude: how iterative design actually works design 21 Learning resources learning