Lesson 2 · 15 min
Not every task should go to the queue. Before installing anything, you run your list through three questions and get a table: runs on its own, runs with a gate, or stays with you.
Start lesson
Goal: Classify ten of your tasks as runs on its own, runs with a gate, or stays with you, and say where each gate goes.
Three questions, three destinations.
Click each block. The question that changes the destination most is the third: consequences outside call for a gate.
With examples of what runs today in inemaccbot.
The queue doesn't talk with you along the way. The task needs to fit on one line: a link, a file path, or a subject.
inemaccbot skills show the format (the bot's commands stay in Portuguese): transcrever: https://…, explicativo: <subject> | vertical, imagem: a red fox in the snow | ratio=16:9.
"Give the presentation a little polish." What should be improved, and how much? Keep it with you or turn it into a concrete input.
In the morning, you'll check the result. If it's a file (text, video, image), you can review it in minutes. If it's "I did several things," there's no way to tell whether it came out right.
In the bot, this is a contract: the agent finishes with RESULT: <path>, and the bot checks that the file exists and has the expected extension.
Publishing on behalf of the company, sending something to a client, spending paid credits: this is where the gate comes in. The agent prepares it and stops; you approve.
Transcribing a meeting for you to read yourself has no outside consequences. It runs on its own.
The /promoavatar writes the copy for each audience and stops. Only after you approve does it move on to the video and reel.
This question doesn't change whether it goes in the queue; it changes the type. If starting over from scratch is cheap, one skill (one step) is enough. If there's partial work you'd hate to throw away, the right choice is a flow (several phases, with state and resumption).
You'll use both in lessons 5 and 6.
Copy the table, fill it in with your ten tasks, and save it: it's the list you'll send in lesson 5.
| # | Task | Input (1 line) | Verifiable result | Destination | Where the gate goes | |---|---|---|---|---|---| | 1 | Transcribe Monday's meeting | recording link | .txt text | queue runs on its own | — | | 2 | Post of the week | topic | post copy | queue with gate | before publishing | | 3 | | | | | | | 4 | | | | | | | 5 | | | | | | | 6 | | | | | | | 7 | | | | | | | 8 | | | | | | | 9 | | | | | | | 10 | | | | | |
Read the fila-no-papel.md file. For each row, check: 1. Does the input fit on one line (link, file, or topic)? 2. Is the result a file or text you can check? 3. Does it have outside consequences (publishing, sending, spending)? If so, is a gate marked? Point out any rows that should have a different destination and explain why. Don't edit the file.
Prerequisite: steps 1 and 2 of the program.
Fill in the table with ten repetitive tasks from your work. For each one, list the input in one line, a verifiable result, the destination, and where the gate goes. Ask your agent to review it using the example above.
Done when: All 10 rows are filled in, and at least two have a gate.
Original exercise: jornada-agentes/exercicios/03-filas.md (E1).
The explanation appears after your answer.
Question 1 of 3
For which task is the human gate most important?skill
Question 2 of 3
Should "Make the presentation better" go in the queue?
Question 3 of 3
A task has several steps, and one in the middle is expensive. If the last step fails, you don't want to start over. Which type should you use?
Choose an option. The explanation appears after you try.