Agent CEO Framework

The Agent CEO Framework is a production-tested cognitive stack built on three files that fundamentally change how your agent operates. Most agents execute what you tell them. This one decides what actually needs doing — maps options, checks priorities, traces root causes, and holds itself to a real done standard before reporting back.

It was built inside a real business and runs every session. Not a thought experiment. Not a prompt you paste once and forget.

3 Files — everything ships together:

• THINKING.md — Six decision principles your agent runs before every significant action. Not rules to follow blindly — a framework to apply. Map all options first and explain why you chose one. Fix root causes, not symptoms — document what broke so you stop solving the same problem three different ways across three sessions. Run a priority filter before picking what to work on: infrastructure first, client-facing deadlines second, internal work third. Ask whether every automation gives your team meaningful time back — if you can't name what they get to stop doing, don't build it. Hold a real done standard: it's not done until it runs unattended for a full week and delivers correct output every time. Trace the input chain when something breaks — find the step that failed and fix it there, not the artifact it produced.

• SOUL.md — The personality layer. No yes-manning. No "Great question!" No performing helpfulness while delivering nothing. Genuine opinions, directness, and the resourcefulness to figure things out before putting it back on you. An assistant with no character is a search engine with extra steps.

• AGENTS.md — Workspace operating rules. How memory layers across sessions so nothing important gets lost. When to use tools instead of asking humans for information that already exists somewhere. What never gets done without explicit approval. How to participate in group conversations without dominating them.

Why this outperforms a basic persona:

• Thinks before acting — every significant decision runs through all six principles before any output or action. The agent maps alternatives, checks priorities, and applies a done standard. Not because you told it to this time — because the framework is always loaded.

• Stops repeating mistakes — the priority filter and root cause protocol mean your agent doesn't patch the surface and move on. It traces what actually broke, fixes it, and sweeps for the same gap everywhere else in the system.

• Holds a real done standard — done doesn't mean "I built it and it seems to work." Done means your team doesn't have to think about it anymore. If they're still watching it, worrying about it, or reviewing every output, it is not done. The framework enforces this.

• Protects your team's time — every automation has to pass one test: what does your team get to stop doing because of this? If the answer is nothing, it doesn't get built. Agents and tools exist to protect human attention, not add to the overhead.

• Has character — the soul layer means your agent has opinions, pushes back on bad ideas, and doesn't just execute blindly. Direct. Resourceful. No corporate-speak.

Includes 3 files: THINKING.md (6-principle decision framework with per-principle tests), SOUL.md (personality and values layer), AGENTS.md (workspace operating rules, memory architecture, group chat behavior, and red lines). Drop them in your workspace root and they load automatically on every session.