🤖 Agents that do the work

Until now, you’ve talked with AI step by step: you ask for something, it responds, then you ask for the next thing. Delegate it’s different — you give it the goal whole at once and let the agent handle everything on its own: it reads, decides, does the work, checks it, and comes back with the finished work.

It’s like asking a human assistant: you don’t dictate every click. You say "organize the invoices folder by month and let me know if anything is missing" and trust that it’ll handle it. Your job becomes describe the goal well e check the result — not monitoring every move.

Aja como um agente que executa a tarefa DO COMEÇO AO FIM.

TAREFA: <ex: organizar estas anotações soltas num resumo
de reunião com decisões e próximos passos>.

MATERIAL: <cole aqui suas anotações / dados>

Como quero o resultado: <ex: um resumo em tópicos, com uma
lista de tarefas no final, cada uma com responsável>.

Trabalhe sozinho até terminar. No fim, me entregue:
1) o resultado pronto,
2) uma linha dizendo como EU posso conferir se ficou certo.

If an agent can already do the whole task on its own, why not have several working at the same time? That’s the fan-out — a fancy word for “spreading out the work.” You split the mission into parts and send one agent to each part, all working in parallel.

Imagine you need to research 5 competitors. Instead of having one agent analyze them one by one (slowly), you send 5 agents, each handling a competitor. They work together, and you get all 5 reports in almost the same time it would take to get one. That's where the speed takes off.

To distribute work among several agents (or just get more out of one agent), first you need to break the task into pieces. A task that’s too big confuses even AI: focus gets lost, details slip through. Smaller pieces are easier to do well—and easier to check.

The image below summarizes the module: you're the conductor coordinating agents that do the heavy lifting, each in their own area.

Delegate no it’s giving up control. The agent does the heavy lifting, but the result has your name on it — so the review is still yours. The good news: reviewing is much faster than doing the work, and with practice you develop an eye for where mistakes tend to show up.

The right review isn’t rereading everything line by line. It’s check the points that matter: was the goal met? Do the numbers add up? Is anything made up? Is anything missing? When you clearly defined “what good looks like” at the top, review is easy—you just compare.

More agents aren’t always better. Launching five when one would do only creates confusion and unnecessary expense. The choice is simple when you look at the task: does it breaks it into independent parts or is one thing, in sequence?

Use the two charts below as your compass. When in doubt, start with one agent—it's easier to keep track of—and only move on to several when you feel you can split them up without the parts depending on one another.

Here, the module’s five ideas become one dance. Orchestrate it’s your job as the conductor: you define the goal, break it into pieces, launch the agents (one or several), review what each one brings back, and combines everything into one result that feels like you.

Notice: you didn’t touch an instrument. You didn’t program or do the heavy lifting. But you conducted it—and the result bears your name. That’s the heart of this entire course, now put into practice with a team of agents.