From prompt to role|INEMA.CLUB-PRO

Lesson 8 · 14 min

The same task in the three eras, and your plan

The best proof that the way of working has changed is to do the same task three times: with just the prompt, with prompt and context, and with the full role. You see with your own eyes what each era delivers. Then you build the plan with dates to put your first agent to work, one step at a time.

Start the lesson
Desk seen from above with a paper calendar, a badge, a pen and a notebook with checked items.

In this lesson you will

  1. 1The same task in three versions
  2. 2What to compare side by side
  3. 3The plan with dates
  4. 4The final test of the course

Goal: compare the same task done in all three eras and walk away with a dated plan for your first agent.

01One task, three versions

Tap or click each version. The task is the same; what changes is what you hand to the AI.

Scroll the page: each step below lights up its part of the diagram.

  1. 1/4

    Prompt only

    You write the request in a temporary chat (no memory), outside the project: no saved rules, no example, no measure. It's the starting point: it shows what the AI does on its own.

  2. 2/4

    With context

    The same request, now inside the project with the base from lesson 3. The tone and the company rules arrive before the request.

  3. 3/4

    Full role

    The sheet from lesson 4, the goal from lesson 5 and the procedure from lesson 6. The agent checks itself against the Measure and hands in the proof along with the work.

  4. 4/4

    The evolution on your desk

    Version 1 shows what the AI does on its own. Version 2 shows what context adds. Version 3 shows what managing adds: a goal, an example and training.

Compare using the same ruler: the Measure on the sheet. It applies to all three versions.

02The three requests and the ruler

Use the task from your sheet. Replace the text between < > with your own.

Version 1 · prompt only (temporary chat, no memory, outside the project)
<the request, the way you would have written it before the course>

(To keep memory from leaking in: in ChatGPT, "Temporary chat" at the top of the screen; in Claude and Gemini, a temporary chat or one with memory turned off.)
Version 2 · with context (inside the project)
Read the project's rules and lessons before you start.
<the same request, in one line>
Version 3 · full role
This is your role: follow the role sheet and the project's standard procedure.
Today's input: <the case>
When you finish, show the check against the Measure.
Comparison (document or spreadsheet)
# Comparison · <task> · <date>
Ruler: <the Measure from the sheet>

| version | passed the Measure? | what got better | what got worse | your time |
|---|---|---|---|---|
| 1 · prompt only | | | | |
| 2 · with context | | | | |
| 3 · full role | | | | |

Conclusion in one sentence: <what you will do from now on>

03Practice: the comparison

The same input in all three versions.

E1 · Three versions

30 min

Run the task from your sheet in all three versions, with the same input, and save the three results.

Done when: you have three result files, one per version, with the same input.

E2 · The ruler

15 min

Fill in the comparison using the Measure from the sheet as your ruler.

Done when: the three rows say "yes" or "no" on the Measure and there is a one-sentence conclusion.

04The course map

The five steps of the course. If you did the practice exercises from lessons 3 to 7, use the plan for your SECOND agent (or to review the first one). Tap or click an exercise to see its "done when".

Week 1Context

rules and lessons

3 exercises

The foundation the agent reads before any request.

See the lesson
Week 1Role sheet

the role

3 exercises

Role sheet checked and the first deliverable.

See the lesson
Week 1Measure

goal

3 exercises

A goal with a number or a test, and a deadline.

See the lesson
Week 2Training

in a cycle

4 exercises

Standard procedure written down.

See the lesson
Week 3Cost

the cost

3 exercises

One week of use logged and one decision tested.

See the lesson
At the end of the courseAn agent with a role, a measurable goal, a trained procedure, and a tracked cost.
16 exercises
Tap or click an exercise to see what to do and the "done when".
  1. 1/6

    Context

    First, the foundation: a rules file, the first lesson from a real mistake, and the new-session test.

  2. 2/6

    Role sheet

    Requires the context foundation. The reference work you did yourself comes before the role sheet.

  3. 3/6

    Measure

    Reinforces the role sheet's Measure: three requests rewritten and the agent showing the check.

  4. 4/6

    Training

    Requires the role sheet with a Measure. Five rounds, the list, and the cycle until it hits zero; the cycle can take two days.

  5. 5/6

    Cost

    Requires the procedure: you use it to test a smaller model or the batch.

  6. 6/6

    The whole path

    Context, role sheet, measure, training, and cost, in that order. Only move up when the "done when" of the previous step is true.

Each step builds on what the previous one set up.

05Generate your plan with dates

Pick a start date and how much time you have each week. The plan respects the exercises that need days of real use.

06The course in three sentences

Stop hunting for the perfect promptDefine the role: a goal you can measure, context, and an example of your own.
Train in cyclesSeveral rounds, a list of corrections and a written procedure. The same mistake doesn't come back.
Watch the costThe right model, cache, batching and a bill you keep an eye on. There are no miracles: you can only relax once the system runs right.

07Final test

Four questions about the whole course.

Question 1 of 4

When you compare the three versions, what is the yardstick?skill

Keep going from here