MODULE 1.3 / 3 OF 9
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.
Area page: eventos.inema.pro/ia-cultivada/en/ · in the menu: “you don’t program it; you cultivate it”
1The process that produces behavior
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.
In practice
Two teams use the same model to answer customers. One has a context page, annotated examples, and written rules. The other just opens the chat. The difference in quality comes from the environment, not the AI.
✓ Do
Write one line about what you already "cultivate" around the AI you use today: a context file, rule, or example?
✗ Avoid
Judging the area by its name: read the main idea on the page before deciding if it’s for you.
2The eight elements of cultivation
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.
In practice
A triage agent does well in its first week, then repeats the same mistakes. It has a function, context, tools, and rules, but no annotated examples, no evaluation, and no feedback that makes it back into the file.
3Each cycle is a new harvest
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.
In practice
An agent that drafts collection emails stays at level 1 for three weeks, with no new entries in the failure log. Only then does it move to sending drafts for a quick review at level 2.
Steps to try
- Open the area page beside this lesson.
- Open a failure log with the page’s four columns and record your agent’s next failure.
- Write one sentence about what changed in your understanding.
4Personal and Jarvis gardens
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.
In practice
Someone uses AI as a spending analyst with a spreadsheet exported from their bank. The rule is: never move money, only show it. After a few weeks, patterns emerge that they had never noticed.
✓ Do
Choose one of the five personal areas and write down the role, one rule, and what you want to evaluate.
✗ Avoid
Jumping to the tool or course without understanding the problem the area addresses.
5The business garden
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.
In practice
A sales team complains that the qualifying agent “broke in production.” The page’s diagnosis points to examples that cover only easy cases. The fix is to include difficult cases and exceptions in the examples.
6The kit: five files and the course
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.
In practice
A small shop owner takes the eight-question assessment, finds that evaluation is her weak spot, and downloads the business kit. She fills out the profile with the generator right in her browser, without anything leaving her computer.
Criteria for reviewing your worksheet
Use this rubric after the lab. Each row asks for evidence; marking a topic as read doesn’t mean the worksheet is complete.
| Criterion | Expected evidence | If it doesn’t meet the criterion |
|---|---|---|
| Main idea | You can describe the area in one sentence that reflects the page. | Reread the top of the page and topic 1. |
| Audience | You can say who the area is for and who it isn’t for. | Return to topic 2 and write an example from your work. |
| Core elements | You can name the area’s core elements. | Use the module diagram as a guide. |
| First step | You chose a small, concrete step. | Copy the first step recommended on the page itself. |
| Starting point | You know which course, kit, or project to open first. | Check topic 6 and the page’s access section. |
| Source | Every statement in the worksheet comes from the page. | Replace your assumptions with what the page says. |
HANDS-ON / ~10 MIN
The first garden
Keep the area page open: https://eventos.inema.pro/ia-cultivada/en/. Use an example from your work, without personal or client data.
Prompt: set up an agent’s garden
Paste it into Claude, ChatGPT, or Codex. Replace the words inside < and > with details about your situation.
Read https://eventos.inema.pro/ia-cultivada/ (the INEMA Cultivated AI area): the eight elements, the cycle, the three levels of autonomy, and the kit’s five files.
My garden: <personal life, Jarvis, or business>
The agent’s function: <one responsibility>
What already exists: <tools, documents, examples I have>
Create the five files as text: 1) agent profile (function, expected result, human owner, does on its own, asks for confirmation, never does, version v1.0, and date); 2) one-page context; 3) annotated example template (good, bad, approved exception, with the reason); 4) failure log header (date | what broke | smallest possible fix | instruction or infrastructure); 5) weekly scorecard. Start at level 1 (proposes, I execute). Where information is missing, write <fill in> instead of making it up.
Completion criterion
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. Keep the worksheet with the main idea, audience, first step, and starting point.
Open the area page ↗Review what you’ve learned
After a month, you say "your AI got better," but the model is the same. What improved, according to the Cultivated AI area?
View suggested answer
The garden: the context, memory, rules, examples, and tools you adjusted in each cycle. During use, the model does not learn on its own; the system around it is what learns.
If your answer was different, return to the relevant topic and describe the difference in one sentence. This check won’t block your progress.
Module summary
- Cultivate, don’t build; level 1 (model) × level 2 (agent); frozen weights; the system is what learns.
- Function; context; tools; rules and limits; examples; memory; evaluation; feedback; traditional software × Cultivated AI.
- Failure log; scorecard; change to the environment; update the profile version; never start at level 3.
- 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.
- “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.
- Agent profile; context; annotated examples; failure log; scorecard; kits; generator; assessment; /cultivar and /revisao-semanal.
Check the source
Pages read on 28/09/2026. Area content changes; the official page takes precedence over this summary.