Lesson 4 · 18 min
Strategy, work architecture, delegation, orchestration, governance, economics, evaluation and evolution. The first four decide what the agent will do; the last four, whether it keeps doing it. In this lesson you give yourself a score from 0 to 3 on each one, with one piece of evidence per score, and choose which course in the program to take next.
Start the lesson
Goal: score yourself from 0 to 3 on the 8 skills of the AI manager, with one piece of evidence for each score, and choose the next course in the program.
Scroll the page: each skill lights up with the question it answers. The first four decide what the agent does; the last four, whether it keeps doing it.
Scroll the page: each step below lights up its part of the diagram.
Question: what result do we want? Result first, tool second. Don't start with the model or the platform. This is the skill of this course (G0): knowing what you want before you turn anything on.
Question: should this process exist this way at all? Architecture of the work: eliminate, simplify, redesign, and only then automate. Step by step, decide what is a person, an agent, a simple automation, or nothing. This is course G2.
Question: what exactly am I handing to the agent? Goal, responsibility, context, tools, deadline, and evidence of completion. Delegating means handing the agent work it can finish and you can check. This is course G3.
Question: one agent or several? One agent or several, in series or in parallel, with a person in the middle, a supervisor agent, or a reviewer. Split only when there is a signal for it. Also in course G3.
Question: does it act on its own, confirm first, or never do it? Permissions, identity, credentials, logging, and traceability. Authority limits before autonomy: what the agent does on its own, what it confirms, and what it never executes. This is course G5.
Question: how much does the result cost? Model, tokens, time, infrastructure: cost per run, per task, and per result. A cheap agent that gets it wrong and needs two tries cost double the price list. This is course G4.
Question: is it good, compared to what? Quality, time, cost, failures, rework, autonomy, human intervention, and business result, always compared to a starting point. An agent gets a performance review. This is course G5.
Question: what did we learn? The company stops asking only "did the task finish?" and starts asking what it learned from it. A tested improvement is promoted; one that made things worse is rolled back. This is LOOP-R, in course G5.
Strategy, architecture, delegation, and orchestration decide what the agent will do. Governance, economics, evaluation, and evolution decide whether it keeps doing it. People, which doesn't show up as a competency, has its own course: G1.
The 8 competencies belong to INEMA's AI Management area. The 0-to-3 scale is a tool for this course, for conversation and for choosing a track, not an exam.
Four steps. Patrícia, from Horizonte (a fictional study company), is the example before you rate yourself.
0: I've never done it and don't know where to start. 1: I understand the question, but haven't applied it yet. 2: I've done it at least once, and I have the file or record. 3: I do it as a routine and can teach someone else.
The scale measures what you do, not what you've read. Having watched a video about governance is a 1; having a reviewed list of permissions is a 2.
Patrícia knows Hora should have a discount cap (a 1 in governance), but the cap doesn't exist in the tool yet.
For every 2 or 3, write down the evidence: a file, a record, a dated decision. If there is no evidence, the score is a 1. It sounds strict, and it is: self-assessment without evidence always comes out higher.
For scores of 0 and 1, write down what is missing. Your track comes from that sentence.
Evaluation: Hora's dashboard exists, but it has no baseline and satisfaction comes from about 6% of customers. A 1, not a 2.
A low score in the first four means the agent does the wrong thing or does it the wrong way. A low score in the last four means nobody knows whether it is doing well, what it costs, or how to improve it.
A manager who only invests in the first half builds agents that nobody can maintain. One who only invests in the second precisely measures work that was badly designed.
In the first half, Patrícia scores 1 in strategy and delegation and 0 in architecture and orchestration. In the second, 1 in governance and evaluation and 0 in economics and evolution. Both halves are weak: 2 of 12 in each.
The program's suggested order is G0, G2, G3, G4, G1, G5. But if a low-scoring competency is causing harm right now in your process, start with its course. Present risk beats suggested order.
Write the reason for your choice in one sentence. Three months from now, redo the self-assessment and compare.
Patrícia chooses G3: Hora gives discounts with no cap and creates large orders with no approval today. G2 comes right after, to redesign the order.
One page. Save it with the date: you will compare against it three months from now.
SELF-ASSESSMENT · AI MANAGER · <name> · <area> · <date>
Scale: 0 never done it · 1 I understand it, haven't applied it · 2 did it at least once (with evidence)
3 do it as a routine and teach it
COMPETENCY | QUESTION | SCORE | EVIDENCE (2-3) or WHAT IS MISSING (0-1) | COURSE
1 Strategy | what result do we want? | <> | <> | G0
2 Architecture | should this process exist? | <> | <> | G2
3 Delegation | what exactly do I hand over? | <> | <> | G3
4 Orchestration | one agent or several? | <> | <> | G3
5 Governance | does it do, confirm or never? | <> | <> | G5
6 Economics | how much does the result cost? | <> | <> | G4
7 Evaluation | is it good, compared to what? | <> | <> | G5
8 Evolution | what did we learn? | <> | <> | G5
SUM 1-4 (what it does): <>/12 · SUM 5-8 (whether it continues): <>/12
COMPETENCY AT HIGHEST RISK NOW: <> because <>
NEXT COURSE: <G1..G5> · REASON: <1 sentence>
REDO ON: <date, 3 months from now>You are a skeptical evaluator reviewing a manager's self-assessment on the 8 AI management competencies. Do not give scores: challenge the ones I gave. Self-assessment: """<paste here>""" Answer, in this order: 1. Which 2 or 3 scores have no concrete evidence (file, record, dated decision)? They should drop to 1. 2. Is any evidence "I read", "I watched" or "I know"? That is a score of 1. 3. Does the choice of next course match the competency at highest risk? 4. What question should I answer before confirming the choice?
Patrícia Lima · sales · Horizonte Distribuidora, fictional study company
| Competency | Score | Evidence or what is missing |
|---|---|---|
| Strategy | 1 | Hora was turned on with "close as many as possible", with no defined result |
| Architecture | 0 | the order follows the same path as before |
| Delegation | 1 | the prompt says "grant whatever discount is needed" |
| Orchestration | 0 | collections by agent are being decided without a design |
| Governance | 1 | knows a cap is missing; the cap does not exist in the tool |
| Economics | 0 | nobody knows the cost per order |
| Evaluation | 1 | dashboard with no starting point |
| Evolution | 0 | no change tested and recorded |
Sums: 2/12 and 2/12. Next course: G3, because a discount with no cap causes harm now.
First assess Patrícia, to calibrate the ruler. Then assess yourself.
Using only what is in the lab (Hora's prompt, the configuration, the conversations and the dashboard), give Patrícia a score from 0 to 3 on each competency, with the evidence or what is missing. Only then open the example next to it and note where you differed.
Done when: all eight scores are given, each with one sentence of evidence or of what is missing taken from a lab piece, and you noted at least one difference from the example and the reason.
Fill in the template for yourself. Run the skeptical prompt and lower the scores it shows have no evidence. Choose the next course in the program based on the competency at highest risk in your process, and put the date to redo this in three months on your calendar.
Done when: you have eight scores, eight pieces of evidence or "what is missing", the two sums, one chosen course (G1 to G5) with the reason in one sentence, and the reassessment date on your calendar.
Lab pieces: WhatsApp agent and Hora dashboard (fictional study company).
4 questions. The explanation appears after your attempt.
Question 1 of 4
Which competency answers the question "should this process still exist?"
Question 2 of 4
You took a course on AI costs, but never calculated the cost per result of an agent. What score in economics?
Question 3 of 4
The last four competencies (governance, economics, evaluation, evolution) decide what?
Question 4 of 4
The suggested order starts with G2, but your agent is giving discounts with no cap today. Where do you go next?
Choose an option. The explanation appears after your attempt.