Delegate a responsibility, not a task
The delegation letter with five parts: objective, what matters, what must not happen, resources, and how to judge the result.
The problem
You're still writing the whole recipe, step by step.
The giant 30-step prompt worked in 2023. With agents, it becomes the bottleneck. Each problem below has a course that addresses it.
- You hand the AI loose tasks and keep doing the work of coordinating everything.
- You explain the same procedure every week because nothing became a skill.
- Either you lock the agent down on everything, or you let it loose too much and find the mistake after it has already cost you.
- The agent has tools but no context, and it moves fast in the wrong direction.
- Nobody defined what a good result is, so nobody knows whether the agent got it right.
After the training
By the end of the training, you can:
Write the trust matrix
What the agent decides on its own, what it proposes for your approval, and what it never touches, with a named owner on each line.
Build an agent that grows lesson by lesson
Identity, standing instructions, memory, tools via MCP, scope of authority, and a trust record that ends in expand, keep, reduce, or shut down.
Redesign the process before automating it
Decide, link by link, what is input, action, rule-based decision, judgment, verification, and exception, and what stays with the human.
Coordinate multiple agents
Subagents with clean context, a model per task, an impartial reviewer, and orchestration patterns in Claude Code and Codex.
Turn what repeats into a skill
From brainstorming to deploy with agents: planning, TDD, systematic debugging, worktrees, and skill creation.
7 courses in order · from basic to advanced
Seven courses, from those who decide to those who build.
The first three are for people who make decisions and don't program. Intent Architecture is the bridge. The last three get into Claude Code, MCP, and skills, for those who want to build.
- 1
Start here
The Super-Agents Have Arrived
Task × responsibility, delegation in five parts in an AI chat, the three levels of autonomy, and the delegation letter with six questions.
6 lessonsOpen course 1 - 2
Hands-on
Super‑Agents: from AI that answers to AI that works
A real responsibility from your area handed to an agent with identity, memory, MCP, autonomy matrix, and trust record.
8 lessonsOpen course 2 - 3
Processes
AI Work Architect
Redesign processes so humans direct and agents execute, deciding what leaves people's desks and what stays.
8 lessonsOpen course 3 - 4
Intent · 4 days
Intention Architecture
The pyramid, the 5 pillars of intent, the intent-to-result system, and the models already operating on intent.
- 5
Tools
Computer Use with GPT‑6 Astra
The four-part prompt contract and five workflows for AI to operate the computer, from search turning into a tool to the phone controlled through the computer.
- 6
Orchestration
Sub‑agents: Claude Code specialists
Four tracks: what a subagent is, how to create one with /agents, model and cost per task, an impartial reviewer, and your own multi-agent system.
- 7
Methodology
Superpowers: development with agents
Ten modules from brainstorming to deploy: planning, TDD with an agent, subagents, systematic debugging, worktrees, and skill creation.
Who it's for
Made for you if
- You already use ChatGPT or Claude and want to stop writing the recipe step by step.
- You look after an area and need to decide what an agent can do on its own and what goes through you.
- You want to hand an entire responsibility to an agent without programming.
- You already delegate in Claude Code and want to coordinate multiple agents and create your own skills.
Not for you yet if
- You already have agents running and need to manage them in the company: see the AI Management area.
- You want to prepare your site to receive agents: see the WebMCP area.
Included in the training
Each course comes with projects you can open today.
Architect of Agents Kit, INEMACCBOT, Content2Video, os-agentes, and other INEMA projects show each piece of the architecture working. They come along, with the videos on the topic.
Projects and kits
OS Coachos-agentesAgent Architect KitINEMACCBOTContent2VideoWebMCP ReadinessFable 5.1 · System PromptSTORM ResearchLong Execution Area
Videos
Why this training
What changes compared with collecting prompts.
Start without coding
The first three courses are for people who make decisions and don't code. Claude Code only comes in at course 6.
An agent that grows with you
In course 2, the same agent gains identity, memory, tools and authority level, lesson by lesson.
Control before power
The autonomy matrix, the off switch and the trust sheet come before you let the agent loose.
The complete architecture
Intent, goal, prompt, skill, context, tools, memory and evals, each piece with its own course.
Questions
Before you start.
Do I need to follow the order?
If you make decisions and don't code, take courses 1, 2 and 3 in that order. If you already build, you can go straight to 6 or 7 and come back to 1 for the mental model.
Do I need to know how to code?
Not for the first three courses. Courses 6 and 7 use Claude Code; course 5 uses GPT-6 Astra operating the computer.
Which tools will I use?
An AI chat in the first courses. Later, MCP, Claude Code, subagents and skills.
How big is the training?
The first three courses add up to 22 lessons (6 + 8 + 8). Intent Architecture has 4 days, Computer Use 2 tracks, Subagents 4 tracks and Superpowers 10 modules.
What does AGI-ready mean?
Being ready to work with agents that are given a responsibility, plan, execute and come back with the result. Your role becomes setting the destination and judging the result.
Your next delegation can be an entire responsibility.
Start with The Super-Agents Have Arrived. In 6 lessons you write the delegation letter and decide how much to hand over to an agent.