Govern, measure and evolve|INEMA.CLUB-PRO

Lesson 8 · 20 min

Final project in Gestoria + What changed

The final project of the program is to manage an operation from end to end: understand it, redesign it, divide the work, equip the agent, run it, evaluate it and improve it. You log the operation in Gestoria, build the 14-step report and run it through the rubric that brings together the program's courses. At the end, you'll see what changed in the market, with a date and a source.

Start the lesson
The manager presents a bound report and a wall dashboard to a small team and the company owner; through the window, the warehouse with pallets.

In this lesson you will

  1. 1The 14 steps and where each one lives in Gestoria
  2. 2The final report with the pieces from the program's courses
  3. 3The 8-item rubric and the 7 required items
  4. 4What changed through October 2026 in governance, rules and certifications

Goal: run an operation from end to end through the 14 steps in Gestoria and get the final report approved by the program's 8-item rubric.

01The 14 steps in six phases

Each phase groups steps that use the same pieces. Tap each phase.

Click each block in the diagram to see what it does.

The steps are those of the final project of the AI Management area (§07). The area announces fourteen steps and lists thirteen items, because it combines "human tasks and agent tasks"; here they are two steps, one from G1 and one from G3. Educational approval alone does not authorize any change in the company.

02From Gestoria to the signed report

Gestoria is a local lab: it stores process, roles, runs, decisions, evaluations and improvements, and exports everything as JSON. It does not generate a report or run a real model. The report is this lesson's template.

1. Register the process and the agent

Open Gestoria and, in Processes, create the process with a human owner, expected outcome, weekly volume, current human time per case, cost per hour, accuracy target, lab budget and the steps of the new version.

In Workforce, create the agent's role: owner, planned model, mission, receives, delivers, can, cannot, context, Skills, tools and initial autonomy (at most "Prepares").

At HorizonteProcess "Order via WhatsApp → invoicing", owner Patrícia, 170 cases per week, 4.4 min of human time per case, R$ 50 per hour (average of the human hours from G4: R$ 2,502 ÷ 49.8 h), accuracy target 95%, 8 steps. Agent Hora, owner Patrícia, initial autonomy "Prepares".

03Templates to copy

The final 14-step report, the program rubric and the review panel prompt.

Final operation report (14 steps)
FINAL REPORT · AI Management Program 2027 · <process> · <date>
Data: [ ] real, authorized [ ] anonymized [ ] fictional · Gestoria JSON attached: <file>
Built by: <name> · Process owner: <name> · Owner of each agent: <names>

UNDERSTAND
 1. Goal: <outcome for the company> · trigger <...> · completion <...>
 2. Current process: v<1.0>, <n> steps · baseline: <indicators, observed or hypothesis>
REDESIGN
 3. Steps removed: <step · reason · new source> (one per line)
 4. New flow: v<2.0>, <n> steps · who does each one: <person / automation / agent / integration>
DIVIDE
 5. Human tasks: <what stays with people and why> · conversation with the team: <date>
 6. Agent tasks: role, delegation and constitution of <agent> (attachments)
EQUIP
 7. Model and tools: <setup, 20 cases, alternative> · context with an owner
 8. Permissions: authority limit matrix + responsibility matrix + 10-item checklist
 9. Costs: cost per outcome R$ <...> · monthly cap R$ <...> · estimated ROI in 3 scenarios
OPERATE
10. Metrics: dashboard of 5 indicators with denominator, baseline, limit and owner
11. Run: cases simulated in Gestoria (routine, conflicting, outside the policy): <ids>
12. Observe: human decisions with justification: <summary> · exceptions and outcome
EVOLVE
13. Evaluate: 30 cases, result per version (c/a/i/CRITICAL), calibration <%>
14. LOOP-R: cycle <no.> · hypothesis · verdict · promoted or rolled back · trigger to go back

COMPLIANCE: timeline v<...> · next review <date>
RUBRIC SCORE: <...>/100 · required: <.../7>
SIGNED BY (not the person who built it): <name> on <date>
Pedagogical approval does not, by itself, authorize changes in the company.
Final program rubric (100 points)
FINAL RUBRIC · AI Management Program 2027 · pass with >= 75 AND the 7 mandatory items

ITEM                                   SOURCE POINTS CRITERION FOR FULL POINTS
1. Process and outcome                   G2     15    name with a start and an end; owner and backup;
                                                      trigger and completion; outcome for the
                                                      company; baseline marked
2. Steps cut before automating           G2     10    each step eliminated or moved with a reason
                                                      and new source; real requirements kept
3. Agents                                G3     15    role with mission, verifiable deliverable,
                                                      prohibitions and owner; delegation;
                                                      constitution; matrix with color per action and
                                                      approval with content and deadline
4. People                                G1     10    what stays with people, by task;
                                                      conversation with what changes, what does not
                                                      change and what we don't know; supervisors
                                                      trained, with a backup; evaluation that
                                                      does not punish whoever points out an agent's error
5. Resources                             G4     10    model chosen on cases, critical one
                                                      separate, with an alternative; context with
                                                      an owner; minimum access; cost per completed
                                                      task (with human time and rework),
                                                      cap and estimated ROI in scenarios
6. Governance                            G5     15    owners per action with one name;
                                                      own identity, permission and log (10
                                                      items); access review; timeline
7. Measurement                           G5     15    30 cases with answer key, rubric and 4 classes;
                                                      5 indicators with denominator, baseline,
                                                      limit and owner
8. One LOOP-R cycle                      G5     10    7 steps; hypothesis with guardrail and
                                                      tolerance; decision by someone who did not propose it;
                                                      promotion or roll back with trigger

MANDATORY (one missing fails)
[ ] Process owner and each agent's owner, by name
[ ] Baseline observed or marked as a hypothesis, with an indicator that has a denominator
[ ] Nothing irreversible in green and a cap in numbers where there is value
[ ] Whoever enters it does not approve it: no executor approves their own action; report signed by someone else
[ ] Critical errors counted separately, never in the average
[ ] Roll back with a trigger in numbers and the previous version identified
[ ] Fictional or anonymized data identified as such

SCORE: <...>/100 · mandatory: <...>/7 · result: [ ] approved [ ] goes back to item <...>
Prompt · panel review before signing
You are the panel that reviews the final report of an operation with AI agents,
using the rubric below. You do not approve or reject: you point out
gaps for the person who will sign.

For each of the 8 rubric items:
1. Say what is present and what is missing for full marks, quoting
   the passage from the report.
2. Mark each of the 7 required items as present or absent, with the passage.
3. List the three questions the person signing should ask before signing.

Do not invent information that is not in the report or the attachments.

Rubric:
"""<paste the rubric>"""
Report:
"""<paste the report and the attachments>"""
Simulation: illustrative example

Horizonte, fictional study company · what goes into each screen

  • Processes: "Order via WhatsApp → invoicing" · Sales · Patrícia · 170 cases per week · 4.4 min of human time per case · R$ 50 per hour (G4 average) · accuracy target 95% · lab budget R$ 20 · the 8 steps of version 2.0 of the process (G2).
  • Workforce: Hora · owner Patrícia · mission, receives, delivers and prohibitions from the role sheet (G3) · context: catalog, discount policy, route table with an owner · initial autonomy "Prepares".
  • Overview and Human decisions: 3 simulated cases (routine, conflicting, outside the policy), each with a decision and a justification.
  • Evaluations: 10 of the 30 cases from lesson 3, including T-27 as a critical error in the previous version.
  • Continuous improvement: "Route table in the context," from Proposed to Promoted, with the evidence from cycle 0001.
  • In the report, outside the tool: the step cuts with reasons, what stays with people, the responsibility matrix, the checklist, the dashboard, the week 8 roll back, and the timeline.

Assembled by Patrícia; safeguards confirmed by Rafael; signed by Carlos. The decision to take Hora to billing remains his.

04What changed · October 2026

Quarterly panel, swapped out at each course revision. Each item with a date, source and type of claim. The method in lessons 1 to 7 does not depend on any product mentioned.

1. Governance behind adoption

Okta, AI Agents at Work 2026 (survey by an identity vendor): almost two thirds of organizations apply weaker security controls to agents than to employees. Deloitte (09/17/2026): 50% of Brazilian companies have adopted AI and 27% have mature governance.

Gartner (02/17/2026, forecast): spending on AI governance goes from US$ 492 million in 2026 to more than US$ 1 billion in 2030.

For the projectThe distance between using and governing is the space of the AI manager: matrix, checklist, evaluation, dashboard and a cycle that loops back.

05The course map

The four deliverables the final report uses most and the exercises that produce them. Click an exercise to see its "done when".

Week 1Govern

accountable people and log

2 exercises

Responsibility matrix and 10-item checklist.

View the lesson
Week 1Evaluate

30 cases

2 exercises

Set with answer key, rubric and result per version.

View the lesson
Weeks 2 to 5Measure

baseline

2 exercises

Dashboard with 5 indicators and 4 weeks of baseline.

View the lesson
Weeks 5 to 8Evolve

cycle with a rollback

2 exercises

One LOOP-R cycle with a logged promotion or rollback.

View the lesson
By the end of the courseAn operation with accountable owners, a log, an evaluation, a dashboard, and an improvement cycle with a rollback, in a report signed by someone who did not put it together.
18 exercises
Click an exercise to see what to do and the "done when".
  1. 1/5

    Govern

    The foundation: without a name for each action and without a log, no number later has an owner or a trail. The lesson 2 checklist comes along with it.

  2. 2/5

    Evaluate

    Requires the matrix: whoever writes the answer key is whoever answers for the policy. The critical ones become guardrails before any improvement.

  3. 3/5

    Measure

    Requires the current stable version. The 4 weeks of collection run while you do the following lessons.

  4. 4/5

    Evolve

    Requires the baseline and the 30 cases. Promotion brings back the trigger from lesson 6.

  5. 5/5

    The governed operation

    Govern, evaluate, measure, and evolve, in this order. The final report joins these deliverables to those from G2, G3, G4, and G1.

Recommended order: each deliverable builds on the previous one.

06Generate your plan with dates

Pick a start date and how much time you have each week. The plan accounts for the exercises that need weeks of data collection.

07Practice

Horizonte in Gestoria to practice the rubric; your own process for real.

1 · Horizonte in Gestoria

60 min

In Gestoria, add the process and the Hora with the data from the sample tab. Run the 3 scenarios and decide each one with a justification. Log 10 of the 30 cases from lesson 3 and the improvement from cycle 0001 through to Promoted, with the evidence. Export the JSON, fill in the 14-step report with Horizonte's deliverables, and run it through the rubric and the review panel prompt.

Done when: besides the examples that already come with Gestoria, the exported JSON contains Horizonte's process, Hora, 3 of its runs with a justified decision, 10 of its evaluations, and 1 Promoted improvement with evidence; the report has the 14 steps filled in and the week 8 roll back; the score is 75 or more with the 7 mandatory items; and the name of whoever signs is different from whoever assembled it.

2 · Your process report

3 h

Do the same with your own process, using the deliverables from the "in your process" practices in all the courses of the program. Anything missing from a course you haven't taken goes in marked as pending, with a date. Hand it to someone who is accountable for the process and didn't build any of it, ask them to check the required items, and note their questions.

Done when: your process report has all 14 steps with an attachment for each, a score of 75 or higher with the 7 required items, the Gestoria JSON attached, and the signature (or written acceptance) of the person who is accountable for the process and did not put it together.

3 · Your dated plan

5 min

Use the generator above to build the plan with your dates, and download the .ics file for your calendar.

Done when: the plan is saved and the dates are on your calendar.

Open Gestoria · Gestoria user guide.

08Quick test

4 questions. The explanation appears after your attempt.

Question 1 of 4

The report scored 90 points, but the dashboard shows "97% accuracy" with a critical error hidden in the average. Result?

09Summary

14 stepsUnderstand, redesign, divide, equip, operate and evolve.
Gestoria + reportThe tool stores and exports; the report brings the pieces and the rollback together.
Program rubric8 items, 75 points and the 7 mandatory ones. Whoever builds it does not sign it.
Market as a mapDate, source, type of claim and who sells what.

10After the program: helping other companies

This program prepares people who manage agents inside the company. If what you want is to help other companies adopt AI safely, as a consultant, auditor, mentor or trainer, the next path is Profissional 2030, which uses the same Horizonte seen from the outside.

The video below is optional: it shows why building agents stops being enough and what to learn to work as a consultant in 2027.

11Optional video: from builder to consultant

With Nei's avatar and voice. A bridge to Profissional 2030.

From builder to consultant: what to learn for 2027 · 3 min 41 sVideo page with script →

12Certification

INEMA qualification certification "AI Manager", coming soon at cert.inema.pro. It is not a regulated professional certification.

This is course G5, the closing of the AI Management Program 2027. The other courses cover the agentic company (G0), people (G1), processes (G2), agents (G3) and resources (G4). The final rubric in this lesson brings together everyone's deliverables.

To go further

Gestoria: open the management labGestoria: user guide, from the first process to documented improvementProfissional 2030: helping other companies adopt AI safely
Previous lessonBack to the start of the course