AI Management Area · the new business management
From now on, the challenge for companies is not only to create agents. It is to know how to manage them: assign role, objective, context, tools, authority and evaluation, just as has always been done with people. And before that, a tougher decision: not to put agents to execute old processes faster. Redesigning the company for the speed of agents. This area brings together the thesis, the 8 competencies of the AI manager, the LOOP-R, and the AI Management 2027 Program: 6 courses published in Portuguese, English, and Spanish, with a guiding case and a final project in Gestoria.
People management + process management + agent management. It’s management, not just technology.
01 · What’s changing
Each era of management responded to a change in what the company had to coordinate. The 2027 era coordinates humans, agents, models, processes and computational budget all at once.
Role, responsibility, tools, limits, evaluation.
Design, document, train, execute. The process as an asset.
Automation: design, automate, monitor.
Prompts, skills, context, tools, memory. Learn to create.
Learn to manage what you built, and redesign work around it.
Any team today can build an agent in an afternoon. What blocks the company is what comes after: who is responsible for it, what it can do on its own, how to measure if it’s good and how much each result costs.
Putting an agent to run a workflow designed for people in 2010 only delivers the error faster. The question that changes everything: if I had humans, agents and automations available today, would this process still exist in this form?
With agents measuring and critiquing their own execution, the process becomes a living system: it executes, observes, proposes improvement, tests, validates, promotes. That’s what INEMA calls LOOP‑R.
On the agent‑building side (prompt, skills, context, tools, memory, evals), see the AGI‑ready area. This page is about the next step: the company that already has agents and needs to manage them.
Open the AGI‑ready area →02 · The thesis
The words change, not the logic. What the company did with a new employee, it now needs to do with each agent: define the role, give context and tools, set limits, evaluate.
Before · people management
employee · role · responsibility · tools · limits · evaluation
Each item existed because without it a person couldn’t work, or worked without control.
Now · agent management
agent · role · objective · context · tools · autonomy · evaluation
The same items, plus two that people didn’t have: explicit context and graduated autonomy. And one detail: the agent can be measured on practically everything.
Separation of duties, authority limits, and human review apply to the agent just as they did to the assistant. The agent goes up to the signing door, never beyond.
First read, never the final word. Responsibility isn’t delegated, it’s organized.
Administration for the AI, people for the team. The limit is a management decision made before the agent is turned on.
Don’t put agents to execute old processes faster.
Redesign the company for the speed of agents.
03 · Agent Culture
In 2025, according to SignalFire’s talent report, an OpenAI engineer was eight times more likely to join Anthropic than the other way around, and Anthropic retained 80% of its hires over two years. In January 2026, the same company published a constitution for Claude: its values, priorities, and the reasons behind each one, used in training. Both cultures want the same thing: someone who exercises judgment, not just obedience.
People management
mission · culture · training · autonomy · performance review · development · leadership
People have needs, rights, and lives of their own. A strong culture is healthy only when it leaves room for disagreement.
Agent Management
role · constitution · examples and test cases · permissions · scorecard · review · human oversight and audit
Today, agents are managed through performance, reliability, and safety. A document given to an agent stays in its context; it doesn’t change the model internally.
One page covering its mission, principles, priorities, limits, when to escalate, and examples of difficult decisions. Don’t copy Claude’s; create a version for your operation.
What the agent handles on its own, what it prepares for a person to approve, and what it sends straight to the manager. Setting the limit is a management decision.
A review that doesn’t improve the score on the same test cases hasn’t improved the agent. This is AI cultivation: shaping behavior gradually instead of configuring it once.
The challenge used to be hiring people with good values and teaching them processes.
Now it’s also about building agents that understand those values and know when to ask for help.
04 · The 8 competencies of AI Management
In the order a manager uses them. The first four decide what the agent will do. The last four decide whether it keeps doing it.
What result do we want?
First the result, then the tool. Don’t start with the model or the platform.
Should this process still exist?
Eliminate, simplify, redesign, and only then automate. That’s the AI Work Architect.
What exactly do I hand over to the agent?
Goal, responsibility, context, tools, deadline, and evidence of completion. It’s the heart of Super‑Agents.
One agent or many? In series or in parallel?
Human in the loop, supervising agent, reviewing agent. The Execution Architect decides parallel vs sequential.
What does it do on its own, what does it confirm, and what does it never execute?
Permissions, identity, credentials, audit, logs, and traceability. Authority before autonomy.
How much does it cost to produce the result?
Model, tokens, time, APIs, Computer Use, infrastructure. Cost per run, per task, per outcome. ROI.
Is it good? Compared to what?
Quality, speed, cost, failures, rework, autonomy rate, human intervention, commercial outcome. The agent has performance evaluation.
What did we learn from this run?
The company stops asking only “did the task finish?”. Enter LOOP‑R.
05 · Process isn’t sacred
The traditional process was designed once and trained. The automated one, monitored. The agentic one, measured. LOOP‑R closes the loop: the execution itself proposes the next improvement.
Before adding AI, break down the process: what exists out of habit, what exists out of a real requirement, what only existed because there was no other way. Automation comes last.
Redesigning the flow means deciding, step by step, what is a human task, what is an agent, what is simple automation, what is Computer Use, what is an API.
A tested and validated improvement is promoted; one that made things worse is reverted. The process has versions, like software. That’s what makes it a living, self‑improving system.
06 · The Manager of 2027
The list of items that land on a manager’s desk has grown. And the cost question has changed: it’s no longer “how much does a call cost?”, it’s “how much does producing the result cost?”.
Where agent economics comes into practice. Per token, frontier models cost more, not less. Economics only appears when measured as cost per completed task: fewer turns, less rework, calibrated effort. A cheap agent that makes a mistake and needs two attempts costs double the listed price. That’s a manager’s calculation, not a programmer’s.
07 · AI Management 2027 Training
People management + process management + agent management, for those who manage and don't need to program. Each course has 8 lessons in INEMA's new format: diagram, steps, a template to copy, practice with "done when" criteria, and a commented test. All of them are published in Portuguese, English, and Spanish.
The third object of management, the 9 objects that cross the manager's desk, the 8 competencies, the market, and the career. It ends with your first diagnosis.
Start with G0 → G2 · Processes Axis · PT · EN · ESChoose the first process, map what really happens, eliminate, simplify, redesign, decide who does each step, and measure the baseline before changing anything.
Open the course → G3 · Agents Axis · PT · EN · ESThe digital role, the delegation sheet, the agent's constitution, the green/yellow/red authority levels, progressive trust, agent team, and exception queue.
Open the course → G4 · Resources Axis · PT · EN · ESCost per result, choosing the model by the work, company knowledge as an asset, tools with minimum access, shadow AI, vendor, and budget with ROI.
Open the course → G1 · People Axis · PT · EN · ESWhat stays with people, fear and resistance, the junior who supervises agents, culture, evaluation, and HR for people and agents. If you work in HR, you can start here.
Open the course → G5 · Governance and evidence · PT · EN · ESWho answers for the agent, permissions and logging, evals, indicators, LOOP-R, promoting and reverting, incoming regulations. It closes with the final project in Gestoria.
Open the course →A distributor with 40 people, from the Profissional 2030 lab. You play the role of Patrícia Lima, a sales manager who has just received agents on the team, and you follow the order that arrives via WhatsApp through to invoicing. The lab's failures become management problems: authority levels, permissions, measurement, shadow AI. Each lesson also applies to one of your own processes.
In the last lesson of G5: understand the objective · analyze the process · eliminate steps · redesign the flow · define human and agent tasks · choose tools and models · define permissions · establish costs · define metrics · execute · observe · evaluate · improve with the LOOP-R. Fourteen steps in the simulator, evaluated by rubric.
Get to know Gestoria →Status of the program as of 10/19/2026. All 6 courses are published in Portuguese, English, and Spanish, open and free. The "AI Manager" certification in inemacert does not exist yet: it is coming soon. It will be an INEMA qualification certification, not a regulated professional certification.
08 · What INEMA already has
Further reading for the program: each course below goes deeper into a competency or a module. The 6 courses of the program are in the previous section.
For decision‑makers: which responsibility to assign to an agent, with authority level, autonomy matrix and trust sheet.
Open the course → What changes · 6 lessonsWhat changes in your work when AI stops waiting for orders. The starting point of Module 1.
Open the course → Work architecture · 8 lessonsRedesign processes for humans to direct and agents to execute. The question “should this process still exist?” originates here.
Open the course → EvolutionExecute, observe, critique, propose, experiment, validate, promote, repeat. The entire Module 10. It also includes the project with guide.
Open the course → Economics and governanceFrom AI‑assisted work to corporate process automation: work, process, agent, automation, Computer Use, metrics and governance. Redesign before automation.
Open the course → OrchestrationOne agent or many, in series or in parallel, with supervisor or reviewer. Competency 4.
Open the guide → Governance by professionLimits, authorities, human review and segregation of duties across three professions. Healthcare · Accounting · Legal.
Open the legal notebook → Final projectThe agentic company management simulator: objectives, agents, processes, outcomes, costs, errors, evaluations, approvals, decisions and improvements.
Open the guide →The operating system for AI‑enabled work: how the company organizes agents, skills and processes in one place.
Open → SpecificationSeven models that become an agent specification on a single page. It is the delegation sheet for Competency 3.
Open the guide → KitSupport materials for the Copilot + Agents for Companies course, ready to use in your company.
Open the guide →09 · Get started now
Take a process that today runs with people, ask whether it should still exist that way, and enter the program through G0. The courses are open. What changes when you join the ecosystem is having people, material and support instead of doing it alone.
The portal with all INEMA courses and projects: tracks organized by level, search by topic and the latest updates. It's where you start.
The community of those applying AI to real businesses: feed curated by Nei with what matters, groups by topic and support material from all trainings.
The platform for those who want to use AI to grow in practice: the more than 400 ecosystem projects, the trainings and the Brain with content organized for reference.
Unsure where to start? Talk to us on Telegram.
10 · Previous versions
The AI Management course was published as a plan: a ten-module syllabus, method, and templates. That plan became the AI Management 2027 Program, with six complete courses. The plan's page remains open and now points to the new courses.