SecondBrain
Ask the Brain
Index/Conceptupdated Fri Jun 12 2026 08:00:00 GMT+0800 (Philippine Standard Time)

Four C's Framework

aiosframeworkclaude-codeagentscontext-engineeringskillscadence

Four C's Framework

Nate Herk (AI Automation)'s architecture for building an AI Operating System (AIOS). Four layers, each of which can't work without the previous one. His later refinement (I Turned Claude Fable Into The Ultimate Second Brain (Nate Herk)): the four split into two tiers — Context + Connections = the second brain (knowledge), Capabilities + Cadence = the AIOS (action) — and it doubles as a teaching sequence for upskilling a team or selling AIOS setups.

C One-liner Test question Maps to
Context It knows your business "What does this business do and who works here?" Context Engineering, CLAUDE.md/memory files
Connections What it can actually touch "What's on my calendar tomorrow? What did John send me?" APIs / MCP servers / local files
Capabilities How you do work "Write a LinkedIn post in my style" Skills (Claude Code)
Cadence Things that happen unattended runs "while your laptop is closed" scheduled jobs / autonomous loops

Layer detail

  • Context — models are stateless; each session reloads global rules + CLAUDE.md + memory/instruction files, else it starts as "a complete beginner." Good output ⇒ good context; a bad answer to "what should I do next week?" means you owe it better context and better hands to pull data. Context is king, not the model. Refinement: context is a routing tree — CLAUDE.md as router, not knowledge dump (see Architecture Engineering).
  • Connections — build the list by auditing your week: where do you go for revenue, customer data, calendar, internal comms, tasks/PM, meetings, and knowledge? Connect one endpoint at a time. (See also CLI vs API vs MCP for how to connect.) Refinement: connections are specifically the live/changing data (emails, P&L, team chat); static data (transcripts, background docs) is plain context and goes visibly stale — his AIOS confidently reported a months-old subscriber count.
  • Capabilities — encoded mostly as skills: reusable instructions for how you specifically do a task. Built forward (skill-creator + iterate) or reverse-engineered from a good end-to-end output.
  • Cadence — the autonomy layer; this is where Bike Method discipline and scoped keys matter most, because unattended runs are where a mis-scoped agent does damage (the 150k-email incident).

Why the dependency order matters

You can't schedule (cadence) a capability that doesn't exist; a capability is useless if the agent can't touch the data (connections); and connections produce generic output without context. Build bottom-up.

Cross-links