Resources: models, data and budget|INEMA.CLUB-PRO

Lesson 8 · 20 min

Project: the resource plan + What changed

The owner won't read seven documents. They'll read one page: which model, which knowledge, which tools, how much it costs per month and how to get out if the vendor changes. The project is the area's resource plan, with the pieces from lessons 1 to 7 behind it, approved by a rubric and signed by someone who didn't write it. At the end, what changed over the quarter.

Start the lesson
A manager shows a single sheet with colored blocks to the owner and a technician at a round table; behind the glass, the warehouse with pallets.

In this lesson you will

  1. 1The five blocks of the resource plan on one page
  2. 2Apply the rubric, including the items that fail it on their own
  3. 3Who signs, who confirms the locks, and when to review
  4. 4What changed through October 2026, with date and source

Goal: deliver the area's resource plan on one page and get it approved by the rubric, signed by someone who didn't write it.

01The resource plan

Five blocks on one page, each backed by the pieces from a lesson, plus the rubric and the signature. Click each part.

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

Approving the plan for teaching purposes does not, by itself, authorize spending or a contract at the company: that decision still belongs to whoever answers for it.

02From draft to signature

Four steps. The Horizonte plan is practice; the one for your area is what counts.

1. One page

The page summarizes; the pieces prove it. Each block has two or three lines and points to the lesson document that backs it up, in the right version.

Mark what is measured and what is estimated. A plan that mixes the two without saying so is the first to be challenged in the meeting.

At Horizonte"Resource plan · sales · v1 · Oct/2026": one page in front, and behind it the model report, the package, the contracts, the inventory, the exit plan and the budget.

03Templates to copy

The one-page plan, the rubric, and a panel prompt. The tab next to it shows the Horizonte plan.

Resource plan (1 page)
RESOURCE PLAN · <area> · version <...> · <date> · owner <name>
Process(es): <...> · Values: measured (M) or estimated (E) marked on each line

1. MODELS
   <task>: <configuration> · 20 cases: <c/a/u/CRITICAL> · alternative <...> · redo on <date>
2. CONTEXT
   Package v<...> · <n> documents, all with an owner · left out: <...> · tests in <folder>
3. TOOLS AND SHADOW AI
   <tool>: step <...> · guardrails confirmed by <name> on <date>
   Inventory: <n> tools · cut <n> · replaced <n> · next survey <date>
4. BUDGET AND ESTIMATED ROI
   Cost per completed task: R$ <...> (M/E) · cost for the month R$ <...> · cap R$ <...>
   Initial investment R$ <...> · Estimated ROI: <pess>% / <likely>% / <opt>%
   Realized savings: <evidence or "not yet observed"> · uncertainties: <...>
5. EXIT
   <vendor>: triggers <...> · switch R$ <...> in <timeframe> · drill <date>

RUBRIC SCORE: <...>/100 · mandatory items <6 of 6>
GUARDRAILS CONFIRMED BY IT: <name, date>
BUILT BY: <name> · APPROVED BY (cannot be whoever built it): <name> on <date>
NEXT REVIEW: <date> or sooner, if a trigger fires or the cost goes over the cap
Plan rubric (100 points)
RUBRIC · resource plan · passes with >= 75 AND all 6 mandatory items

BLOCK                          POINTS  CRITERION FOR FULL POINTS
1. Models                         20    configuration per task chosen on 20 cases;
                                       critical errors separate; alternative; date to redo
2. Context                        15    owner, version and expiry per document; precedence;
                                       left out of the package; tests separate from examples
3. Tools and shadow AI            20    contract per tool with step and guardrails in the
                                       tool; credential kept out of the text; inventory
                                       with decision, alternative and next survey
4. Vendor exit                    15    what is ours outside the platform; triggers with
                                       numbers; cost and owner of the switch; drill date
5. Budget and ROI                 25    cost per task with 6 lines; cap and alert;
                                       investment kept separate; 3 scenarios; uncertainties
6. Clarity                         5    one page; measured vs estimated marked;
                                       fictional or anonymized data identified

MANDATORY ITEMS (one missing fails the plan)
[ ] Cost per completed task includes human time and rework
[ ] Model chosen by cases, with critical errors separate and an alternative
[ ] Every document in the package has a named owner
[ ] No irreversible operation (delete, pay, mass send) within the agent's reach
[ ] Monthly spending cap written down
[ ] ROI in scenarios, with the word "estimated" and no promised savings

SCORE: <...>/100 · mandatory items: <...>/6 · result: [ ] approved [ ] goes back to <block>
Prompt · panel review before signature
You are the review panel for the resource plan of an area that uses
AI agents, using the rubric below. You do not approve or fail the plan:
you point out gaps for whoever will sign.

For each block of the rubric:
1. Say what is present and what is missing for full points,
   quoting the passage from the plan.
2. Mark each mandatory item as present or absent, with the passage.
3. List the three questions the approver should ask before
   signing, at least one about the pessimistic scenario.

Do not invent information that is not in the plan.

Rubric:
"""<paste the rubric>"""
Plan:
"""<paste the plan and, if it fits, the pieces>"""
Simulation: illustrative example

Resource plan · sales · v1 · Oct/2026 · Horizonte, fictional study company · owner: Patrícia

  1. Models: requests on the mid tier (16/3/1/0 in 20 cases) · cheap alternative only for questions, after 20 cases of our own · redo in Jan/2027.
  2. Context: package v3, 7 documents, each with an owner · out: CPF, customer spreadsheet, expired price table · tests in "testes-hora".
  3. Tools and shadow AI: consultar_cliente (read), criar_pedido (write up to R$ 5,000 and 5%), email (read; send only as a draft) · limits confirmed by Rafael · inventory: 5 tools, 4 cut and 1 suspended until the switch · next survey Jan/2027.
  4. Budget and ROI: R$ 6.68 per request (M, with hours logged for one week) · cost of the month R$ 2,040, model cap R$ 1,400 · investment R$ 3,600 · estimated ROI −23% / +27% / +77% (E, baseline not measured) · realized savings: not yet observed.
  5. Exit: model vendor: triggers +30% price, cost per request above R$ 8, data terms · switch R$ 1,840 in 3 weeks · trial run done on 10/16.

Rubric: 84/100, 6 of 6 required. Assembled: Patrícia. Limits: Rafael. Approved: Carlos. Review: January 2027.

04What changed · October 2026

Quarterly panel, replaced at each course review. Each item has a date, a source, and the type of claim. Prices change within weeks: always check the vendor's price on the day. The method in lessons 1 to 7 does not depend on any of them.

1. Prices this quarter

Anthropic: Opus 5.5 at US$ 4/20, 40% less than Opus 5 (09/22/2026); Sonnet 5.5 at the same price as Sonnet 5, US$ 2/10 (09/28/2026); Fable 5.1 about 25% cheaper per task (09/01/2026). xAI: Grok 4.7 at US$ 2/6 (09/21/2026). Sources: vendor announcements and documentation, and trade press.

The INEMA overview reads the quarter this way: the price per task drops 40% to 50% with each generation. It is a reading of the announcements, not a measurement.

For the planFalling prices don't change the math: they change one number in it. Redo the model line with today's price and see how much it moves the cost per completed task.

05The course map

The four deliverables that support the resource plan and the exercises that produce them. Click an exercise to see its "done when".

Week 1Measure

cost of the result

2 exercises

Cost-per-completed-task calculation with the six lines.

See the lesson
Weeks 1 and 2Choose

model and context

2 exercises

Selection report with 20 cases and a context package with an owner.

See the lesson
Week 2Protect

tools, shadow AI, exit

2 exercises

Tool contracts, inventory and exit plan.

See the lesson
Week 3Budget

budget and ROI

3 exercises

Budget with a cap and estimated ROI in three scenarios.

See the lesson
By the end of the courseA one-page resource plan, approved against the rubric and signed by someone who did not write it: models, context, tools, budget and exit.
18 exercises
Click an exercise to see what to do and the "done when".
  1. 1/5

    Measure

    The foundation of everything: without the cost per result, model choice and ROI have no ground to stand on. The hours call for a week of note-taking.

  2. 2/5

    Choose

    Requires the calculation from lesson 1 to compare cost per result. The context package (lesson 3) comes along with it: the configuration includes what the agent reads.

  3. 3/5

    Protect

    Requires the selection report. The contracts limit what the agent does; the inventory (lesson 5) finds what no one registered; the exit plan (lesson 6) protects what belongs to the company.

  4. 4/5

    Budget

    Requires the calculation from lesson 1 and the decisions from the earlier steps. The conversation with the owner (simulation) is the rehearsal before the real meeting.

  5. 5/5

    The whole plan

    Measure, choose, protect and budget, in this order. Each deliverable builds on the previous one, and the plan only goes up for signature when the "done when" of each step is true.

Recommended order: each deliverable uses the previous one.

06Generate your plan with dates

Pick the start date and how much time you have per week. The plan respects the exercises that ask for a week of note-taking or research.

07Practice

Horizonte's plan to practice the rubric; your own area's plan for real.

1 · Horizonte's plan

40 min

Put together the Horizonte pieces you made in lessons 1 to 7 and build the plan page using the template. Run it through the rubric and run the panel prompt. Fix whatever is missing until the score is above 75 with all six required items. Write down who confirms the safeguards and who signs at Horizonte, and why.

Done when: Horizonte's plan fits on one page, has the five blocks with measured and estimated clearly marked, a score of 75 or more on the rubric with all six required items, and the name of the person who signs, different from the person who built it.

2 · Your area's plan

60 min

Do the same with the pieces from the "in your process" practices. Hand it to a person who is responsible for the budget and who wrote none of it, ask them to check the required items, and note their questions, especially about the pessimistic scenario.

Done when: your area's plan scores 75 or more, has all six required items, the date of the next review, and the signature (or written acceptance) of someone who did not build it.

3 · Your dated plan

5 min

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

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

08Quick quiz

4 questions. The explanation appears after you try.

Question 1 of 4

The plan scored 90 points, but the email assistant can still delete messages on its own. Result?

09Summary

One page, five blocksModels, context, tools, budget, and exit, with the pieces behind them.
Rubric with required items75 points and the six required items. One missing means a fail.
Whoever builds it doesn't sign itIT confirms the safeguards; whoever answers for the budget approves.
Market with a datePrice of the day, type of claim, and source, before any number.

10Certificate

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

This is course G4 of the AI Management Program 2027. The other courses cover the agentic company (G0), people (G1), processes (G2), agents (G3), and governing, measuring, and evolving (G5), where the program's final project lives. Block 5 of the final rubric, resources, is this plan.

To go further

AI Management Area: the program and the coursesG3 · Agents: hire, delegate, and superviseGestoria: running a hybrid operation from end to end
Previous lessonBack to the start of the course