Big Tasks Are Hard
Imagine you sit down and someone says: build me a website, write me a novel, or plan the school science fair. Where do you even start? The goal is real, the finish line is clear — but the path from where you are to where you need to be is completely fogged in. That feeling of being overwhelmed by a big task is not a personal weakness. It is a fundamental challenge that affects humans, teams, and AI agents alike.
What Makes a Task Big?
A task is big when it has three characteristics that make it impossible to complete in a single action. First, it has multiple parts that depend on each other — you cannot decorate a room before you paint it. Second, it takes more time than a single burst of effort can cover. Third, it involves uncertainty — you do not know every step you will need until you are already partway through. When a task has all three of these properties, no single instruction or action will finish it. You must plan.
A task is big when it has multiple interdependent parts, spans more time than one action covers, and involves uncertainty about what steps lie ahead. Big tasks require planning, not just doing.
Think about how an AI assistant might fail if it tried to plan a birthday party in one shot. It might immediately order a cake — but forget to book a venue first, so the cake arrives at a location nobody can use. Or it might invite all the guests before confirming the date with the birthday person. Each single action made sense in isolation, but without a plan connecting the actions together, the whole effort collapses. This is not a problem with the AI's intelligence. It is a problem with attempting a big task as if it were a small one.
The Single-Step Illusion
Humans are remarkably good at making complex tasks look simple. When you watch an expert chef prepare a five-course meal, their hands move fluidly from one action to the next. It looks like one smooth operation. But behind that smoothness is a detailed mental plan built up over years of training — which dish goes in which oven at what time, which ingredients prep fastest, which steps cannot start until another finishes. Beginners see the finished result and try to imitate the surface — and fail — because they are missing the invisible plan underneath. AI agents face the same trap. If an agent is only designed to pick the next best action without any sense of the larger plan, it will produce the AI equivalent of a beginner's kitchen disaster.
An agent that picks the locally best next action, without any larger plan, often succeeds at each step but fails at the overall goal. Individual good moves do not automatically add up to a good strategy.
Match each challenge of big tasks to the reason it makes single-step action fail.
Terms
Definitions
Drag terms onto their definitions, or click a term then click a definition to match.
Why Agents Need to Plan
An AI agent is a program that perceives its environment, makes decisions, and takes actions to reach a goal. Simple agents handle simple tasks: answer one question, move one step, flip one switch. But as soon as the goal is complex, the agent must do something more sophisticated than just react to the current moment. It must look ahead. Looking ahead means constructing a plan: a sequence of steps ordered so that each one sets up the next. Without a plan, an agent is like a driver who only looks at the road directly in front of the hood — they might stay on the road for a while, but they will miss the exit that was only visible half a mile back.
You rarely see a plan — you see the results of a plan. The smoothness of an expert's work, the reliability of a well-run project, the coherence of a long document: all of these are evidence of planning that happened before any visible action began.
Which property makes a task too large to complete in a single action?
An AI agent tries to write a full research report by generating one massive output in a single step. What is most likely to go wrong?
Map the Hidden Steps
- Step 1: Pick one of these big tasks: plan a camping trip, write and publish a school newspaper, or build a simple app.
- Step 2: Without worrying about order yet, brainstorm every single thing that would need to happen for the task to be complete. Write each as a short phrase. Aim for at least 15 items.
- Step 3: Look at your list. Circle any item that cannot start until another item on your list is finished first.
- Step 4: Count how many arrows of dependency you found. Write one sentence explaining why this task cannot be done in a single action.
- Step 5: Reflect: how is your list similar to what an AI agent would need to figure out before it could tackle the same goal?