ai-os-systems-architect

What it does

A structured design-then-build process for AI Operating Systems: agentic systems with a kernel (orchestrator), memory subsystem, process layer (agents/tools), and I/O system (triggers/outputs), mapped directly onto how a traditional operating system works, so the architecture is legible instead of a pile of connected tools.

How it works

Two phases. Design walks through four questions (what the system does, what triggers it, what it produces, how autonomous it should be), then designs each layer in turn: the Kernel (which model, what it always knows, how it decides: linear, branching, or reflective), Memory (working, short-term, long-term, semantic, not every system needs all four tiers), the Process Layer (which jobs need an AI sub-agent vs. a deterministic tool vs. an MCP server), and the I/O System (triggers: manual, webhook, scheduled, event-driven; outputs: messages, database records, files, API calls). All four layers compile into a single Architecture Brief.

Build then walks that brief into an eight-layer implementation sequence, scaffold, kernel core, memory, process layer, I/O, orchestration, testing, deployment, one step at a time, Claude Code prompt by Claude Code prompt.

Try it

I want to design an AI OS for [what it does]. Walk me through the four layers using the
traditional-OS analogy: what's my Kernel (the core orchestrator, what model, what it
always knows, how it decides what to do next), what's my Memory (what does it need to
remember across sessions, and at what tier), what's my Process Layer (which jobs need an
AI agent vs. a plain tool call), and what's my I/O System (what triggers it, what it
produces). Compile it into one Architecture Brief I can hand to a build session.

When to reach for it

Before building anything with more than one moving agentic part: a system that needs to remember things across sessions, react to more than one kind of trigger, or coordinate more than one tool.