LEARNING PATH 01 / INEMA AREAS

Manage and direct agents

AI Management, AGI-ready, and Cultivated AI: work shifts to directing, cultivating, and evaluating agents.

0% 0 of 0

Three areas, with the same deliverables for each: the worksheet and the first step.

3 areas

18 topics

60 min with practice

1 worksheet per area

Learning path map

1.1 ~20 min

🧭 AI Management: 2027, the year of managing agents

2027: managing agents

1.2 ~20 min

🧭 AGI-ready: Your Work Becomes Managing Agents

directing agents

1.3 ~20 min

🧭 Cultivated AI: you cultivate an agent; you don’t program one

you don’t program it; you cultivate it

Detailed content

MODULE 1.1

AI Management: 2027, the year of managing agents

Explain why companies with agents need management, not just technology, using the 8 competencies and LOOP-R.

0% 0 of 0

What it is

The INEMA.events.inema.pro AI Management area opens with a statement: "In 2026, we taught you to build agents. 2027 is the year to learn how to manage them." An agent is an AI that takes on a responsibility and does work with tools instead of just answering a question. Managing an agent means giving it a role, goal, context, tools, authority limits, and evaluation, as you have always done with people. The page also makes a tougher point: don’t put agents to work running old processes faster. Redesign the company for the speed of agents. The area brings together the thesis, the 8 AI manager competencies, LOOP-R, the 2027 AI Management Training, and the INEMA courses and projects that already cover most of the path.

Why learn it

Building an agent is no longer the bottleneck. What holds a company back is what comes next: who is responsible for it, what it can do on its own, how it is measured, and how much it costs. Understanding the thesis changes the question from "which tool?" to "how do I manage this?"

Key concepts

People management + process management + agent management; management, not just technology; redesign before accelerating.

What it is

The page shows the eras of management, each shaped by what companies needed to coordinate. From 1980 to 2000: people management—roles, responsibilities, tools, limits, and evaluation. From 2000 to 2020: process management—design, document, train, and execute. From 2020 to 2025: digital transformation—design, automate, and monitor. From 2025 to 2026: agent building, with prompts, skills, context, tools, and memory. In 2027, management coordinates people, agents, models, processes, and computing budgets at the same time. This area is for companies that already have agents and need to manage them; building the agents themselves is covered in the AGI-ready area.

Why learn it

Treating an agent as an IT project leaves out the part that determines the outcome: responsibility, limits, and measurement. Seeing the agent as the third object of management puts it on the agenda for the people who run the business.

Key concepts

Three shifts: building got easy, managing didn’t; the old process doesn’t deserve to be sped up; processes are no longer frozen.

What it is

The page’s central thesis is that managing agents is a lot like traditional people management, but with more measurement. The words change, not the logic: employee, position, responsibility, tools, limits, and evaluation become agent, role, goal, context, tools, autonomy, and evaluation. Two elements are added that people didn’t have: explicit context and graduated autonomy. Graduated autonomy means the agent doesn’t get "all or nothing" authority. It gets a defined scope—a limit on what it can decide on its own. And you can measure almost everything an agent does.

Why learn it

Many rules companies already know still apply, which shortens the path. Separation of duties, authority limits, and human review don’t need to be invented for agents; they need to be applied to them.

Key concepts

Accounting: the person who enters items doesn’t approve them; legal: the agent prepares, the lawyer decides and signs; healthcare: some decisions are outside the agent’s authority.

What it is

The page lists eight competencies in the order a manager uses them. Strategy asks what outcome we want; work architecture asks whether the process should still exist; delegation asks exactly what to give the agent; orchestration asks whether to use one agent or several, in sequence or in parallel. Governance decides what it does on its own, confirms first, or never executes; economics asks how much it costs to produce the outcome; evaluation asks whether it’s good and what it’s compared with; evolution asks what we learned from this run. The first four determine what the agent will do; the last four determine whether it keeps doing it. The 2027 manager oversees nine objects, from people and agents to computing budgets, and tracks metrics such as cost per outcome, rework, and autonomy rate.

Why learn it

The competencies work as a checklist: if one has no answer, that’s where the agent will fail. They also show you where to study, because the page points to the course that currently teaches each competency.

Key concepts

Strategy; architecture; delegation; orchestration; governance; economics; evaluation; evolution; cost per completed task.

What it is

The page compares four generations of processes. Traditional processes were designed, documented, taught, and executed. Automated processes were designed, automated, and monitored. Agentic processes start with a goal; the agent executes and measures, then a person or agent evaluates. LOOP-R closes the cycle: execute, observe, critique, propose, experiment, validate, promote, and repeat. The process becomes a living system with versions, like software. There are three rules: don’t automate a bad process; people direct, agents execute; promoting and reverting are steps, not accidents.

Why learn it

Without the cycle, a company only asks "is the task done?" and learns nothing from each run. With it, each run suggests the next improvement, and a change that made the result worse can be rolled back without drama.

Key concepts

Four generations; LOOP-R; versioned process; promote and roll back; take it apart before automating.

What it is

The 2027 AI Management Training has ten modules and a final project, with the Gestão de IA course as its backbone. The modules cover the agentic company, process is not sacred, delegation and autonomy, agent economics and governance, and finally the company that learns through LOOP-R. The final project is Gestoria, a simulator for managing an agentic company in fourteen steps. The page is honest: the training packages a thesis; it is not a new course published today, and 75% to 80% of its foundation already exists in open courses and projects. This area brings together the Gestão de IA, Os Super-Agentes Chegaram, Super-Agentes na prática, Arquiteto de Trabalho com IA, LOOP-R, and Copilot + Agentes courses, as well as the Gestoria, LOOP-R, Arquiteto de Execução, Agentic OS, Kit do Arquiteto de Agentes, and Kit Copilot + Agentes projects.

Why learn it

Knowing what is already available keeps you from waiting for a course that has yet to be packaged. The suggested path is to start with the existing courses, following the module order.

Key concepts

Ten modules; Gestoria as the final project; 75% to 80% of the foundation already exists; the best first module is a process from your company.

View full version →

MODULE 1.2

AGI-ready: Your Work Becomes Managing Agents

Explain the difference between giving step-by-step instructions and delegating to an agent, using the ladder of intention, objective, prompt, skill, context, tools, memory, permissions, and evals.

0% 0 of 0

What it is

The AGI-ready area opens with the line: “AI has stopped waiting for instructions. Your work becomes managing agents.” The page recalls how, for three years, we tried to put the entire operation into one giant prompt. With agents that work for hours or days, the game changes: you state the destination, and the agent finds the way. An agent here is an AI that takes on a responsibility, plans, executes, monitors, and returns with the result. This area brings together what is changing, the mental model to keep up, and INEMA courses and projects to help you get started today, without programming.

Why learn it

If you keep operating AI only through short commands, you compete with the agent itself. If you learn to delegate, you multiply your own capacity, and you can practice that.

Key concepts

Destination, not route; work for hours or days; delegate; manage agents without programming.

What it is

The page describes six changes: the first three come from people inside the labs, and the next three are what those changes mean at your desk. At your desk, the first is that a task becomes a responsibility: you stop asking “make this spreadsheet” and start handing over “take care of this month’s reconciliation.” The second is that trust becomes a curve, not a switch: what the agent decides on its own, what it suggests for you to approve, and what it never touches, with an owner for each row. The third is that your value shifts to what AI cannot decide: setting the objective, choosing what matters, stating what must not happen, and judging whether the result is good. The page also cites, as an external source, an English-language video about the article “Alien Minds” by Jakub Pachocki, OpenAI’s chief scientist, who acknowledges that capability is growing faster than control.

Why learn it

Without this shift in how you define the work, you keep writing recipes for a tool that can already take care of the result. Without a trust curve, you either lock everything down or give it too much freedom.

Key concepts

Task × responsibility; trust as a curve; trust matrix; role of an agent manager.

What it is

The page warns that talking about the “end of the prompt” is an overstatement: the prompt is not disappearing. What is ending is the prompt as the entire program, with all the operation’s intelligence written in natural language, as people did between 2023 and 2025. Micro-prompting means teaching every step: do A, then B, then C. If one step is missing, the result is wrong. Now the prompt states the objective, what matters, the constraints, the resources, and what counts as a good result. It works alongside the Skill, which is reusable knowledge or a procedure: instructions, reference files, templates, scripts, and guidance on using tools.

Why learn it

Separating the prompt from the Skill keeps you from rewriting the same nine steps every time. The prompt stays short, and quality comes from the operating manual the agent already has.

Key concepts

Prompt = what I want now (the brief); Skill = how our organization usually does it; Prompt → Context → Skill → Agent Engineering → Orchestration.

What it is

The page lays out the architecture in levels, using one example throughout: increasing course conversions. Intention answers “why?”; objective answers “what?”; the prompt guides this run; the Skill explains how to do the work well, consistently. Next come context (what it needs to know), tools (the agent’s hands, such as a browser, CRM, spreadsheets, APIs, and MCP), memory (accumulated experience), and finally permissions and evals. Evals are tests that show whether the result meets the success criteria. Along with limits, human approval, and logs, they form quality control. As the page puts it, we do not want only a powerful agent; we want an agent we can control.

Why learn it

When an agent fails, the ladder shows which rung has the problem: missing context, a missing Skill, or missing criteria. Without it, every failure leads to a longer prompt.

Key concepts

Intention; objective; prompt; skill; context; tools; memory; permissions and evals; a human sets the destination and judges the result.

What it is

The page turns the change into six tips that fit into your schedule. First: separate tasks from responsibilities in your list of requests to AI. Second: rewrite a prompt as a delegation, with the goal, what matters, what must not happen, resources, and how you will judge the result. Ask for a plan before any action. Third: if you have explained the same procedure three times, turn it into a Skill with a manual, template, checklist, and examples. Fourth: write the trust matrix; fifth: give context before tools; sixth: define what "good" means before you run it.

Why learn it

These are small, verifiable steps, not a months-long project. Each one is the first step toward a skill the page lists: thinking in processes, directing AI, building agents, integrating systems, evaluating and supervising, and understanding the business.

Key concepts

Task × responsibility; delegation in five parts; a Skill for what repeats; trust matrix; context before tools; evaluation before running.

What it is

This area brings together courses for people who make decisions but don’t code, and courses for people who want to build. To start: Os Super-Agentes Chegaram (6 lessons), Super-Agentes na prática (8 lessons), and Arquiteto de Trabalho com IA (8 lessons). To work with the tools: Arquitetura de Intenção, Computer Use com o GPT-6 Astra, Subagentes e Superpowers, and the recording Agentes do Zero: Fundamentos (5 days). Projects with code and a guide include OS Coach, os-agentes, Kit do Arquiteto de Agentes, INEMACCBOT, Content2Video, WebMCP Readiness, Fable 5.1 · Prompt de Sistema, and STORM Research. The page ends with three doors into the same ecosystem: INEMA.CLUB (open courses), INEMA.VIP (community), and INEMA.PRO (hands-on platform).

Why learn it

Each project is a working piece of the architecture: INEMACCBOT shows the trust matrix in code, while WebMCP Readiness shows the "tools" side from the website’s perspective. Seeing a finished example helps you make your first delegation sooner.

Key concepts

Start with Os Super-Agentes Chegaram; open courses in Portuguese; projects with code and a guide; INEMA.CLUB, INEMA.VIP, INEMA.PRO.

View full version →

MODULE 1.3

Cultivated AI: you cultivate an agent; you don’t program one

Explain why an agent improves through the environment you cultivate, not through the model, using the eight elements, the cycle, and the kit’s five files.

0% 0 of 0

What it is

The Cultivated AI area opens with this idea: "Before, we programmed behavior. Now, we program the process that produces behavior." No one writes line by line the ability of a large model to draw analogies or plan a business. Labs create the conditions, and the abilities emerge. The page separates cultivation into two levels. At level 1, the lab cultivates the model through its architecture, data, objective, compute, training, and feedback. You don’t take part. At level 2, you cultivate the agent: the model arrives ready, with frozen weights, and you cultivate the environment around it.

Why learn it

This distinction explains why the same model becomes a confused intern for one person and a reliable professional for another. Your competitor has the same model; the garden is different.

Key concepts

Cultivate, don’t build; level 1 (model) × level 2 (agent); frozen weights; the system is what learns.

What it is

Every agent, whether a personal assistant or a sales agent, is cultivated with the same eight ingredients. Function: what it is responsible for, one role at a time. Context: what it needs to know; one good page is worth more than twenty pages dumped in. Tools: what it can operate, starting with read access. Rules and limits: what it does without asking, what it asks you to confirm, and what it never does. Examples are annotated real cases, including bad ones. Memory is what carries over between sessions. Evaluation asks whether it is good, and compared with what. Feedback turns each failure into a change to the file, not the conversation. The first four make the agent work; the last four help it improve.

Why learn it

The page observes that most people stop at the fourth element and complain that "the AI doesn’t learn." Knowing all eight helps you see what is missing.

Key concepts

Function; context; tools; rules and limits; examples; memory; evaluation; feedback; traditional software × Cultivated AI.

What it is

The cycle is at the heart of the method: process, agent, execution, result, evaluation, feedback, and finally a better agent. What changes from one harvest to the next is not the model. It is what you recorded about where it went wrong and what you adjusted in the environment. The page gives three rules. Evaluation produces log entries, not impressions: one line for each failure, and after ten lines, a pattern appears. Fix it in the file, not the conversation. Autonomy is earned one clean cycle at a time, across three levels: the agent proposes and you execute; it executes and you review; it executes and reports.

Why learn it

Without the cycle, the first four elements produce an agent that stalls. The autonomy rule protects you from the costliest mistake: trusting too much too soon.

Key concepts

Failure log; scorecard; change to the environment; update the profile version; never start at level 3.

What it is

The page calls the places where cultivation happens gardens. In your personal life, three or four text files and a weekly habit are enough. You don’t need a CRM or API. The work can cover five areas: decisions, health and routine, money, learning, and writing and communication. The habit that supports it all is a 20-minute weekly review: read the memory, write three lines in the failure log, update the file, and delete anything that is no longer true. “Jarvis” is the common name for an agentic personal assistant. It doesn’t just answer; it acts in your environment: reading and writing files, working with your calendar, and sending messages. The tool comes ready to use. Jarvis doesn’t; you cultivate it with context, memory, tools, rules, skills, evaluation, and feedback.

Why learn it

Personal agents have access to your life, and the page devotes a section to security, the part no one cultivates. Without a fence, the garden is exposed.

Key concepts

20-minute weekly review; proposes, executes and reviews, executes and reports; credentials in one place; external content is data, not instructions; backup before destructive operations.

What it is

For businesses, the page proposes that you stop treating AI as software to program and start treating it as a capability the organization develops. Rigid automation says, “if A happens, do B, then C,” and breaks at the first case no one anticipated. A cultivated agent gets a work environment with a role, criteria, examples, limits, and an expected result. The first agent always starts at the “proposes, human executes” level: one process, one agent, one metric. The page suggests starting points by area, such as lead qualification in sales, triage in customer support, and reconciliation in finance. It also lists why projects fail, with the smallest fix for each symptom.

Why learn it

When projects fail, the cause is rarely the model. It’s usually a lack of context, process, or feedback. Knowing how to diagnose the problem keeps you from switching tools when the real issue is cultivation.

Key concepts

“The agent hallucinates”: missing or cluttered context; “no one trusts it”: no sample-based evaluation; “it stopped improving”: feedback isn’t fed back into the environment; emerging role: agent manager.

What it is

Any cultivation effort fits into five text files, used in both the personal and business garden. The agent profile describes its function, expected result, human owner, what it does on its own, what requires confirmation, and what it never does, along with the version and date. The context file is “About me” or “About the company.” The annotated examples collect approved good, bad, and exceptional cases, with the reasons why. The failure log has one line per failure. The scorecard records the review and the decision to keep the agent at its level, move it up, or move it back. This area includes the Cultivated AI course, on a single page and also in English and Spanish; the personal, Jarvis, and business kits; the profile generator; the eight-question maturity assessment; and the iacultivada repository on GitHub, with the /cultivar and /revisao-semanal skills. The page points to the AI Management area for managing agents at your company and to the AGI-ready area for building them.

Why learn it

With the files ready, you don’t have to start from scratch or rely on remembering what you agreed with AI. The page sums up the starting point: choose a garden, a role, and three rules. The rest is a cycle.

Key concepts

Agent profile; context; annotated examples; failure log; scorecard; kits; generator; assessment; /cultivar and /revisao-semanal.

View full version →