Give AI Tasks, Not Directions: One Principle That Changes How You Work With AI

Why a Three-Hour Conversation With AI Leaves You With Nothing

Let’s start with an experience a lot of people have had but can’t quite put into words: you open AI meaning to solve one problem, and end up getting more and more into it, one thread leading to the next, three hours gone, feeling like you’ve talked a lot and learned a lot — but when you actually try to produce something concrete, you find your hands empty. Nothing landed.

Someone who uses AI heavily nailed the root cause of this in one sentence:

Why does talking to AI, no matter how much, accomplish nothing? Because what I give AI is all “direction,” never “task.”

This might be the single most important habit to build when using AI — and one of the least explicitly stated.

Direction vs. Task: The Difference Is “Completion State”

The fundamental difference between direction and task comes down to one thing: completion state (Definition of Done).

• Direction: an open-ended topic with no clear endpoint. Things like “help me think through how to approach this topic,” “what should I do about this,” “help me improve this article.” It has no completion state — what counts as “figured out”? What counts as “improved”? There’s no standard, so it can be discussed forever.
• Task: a concrete action with clear boundaries and acceptance criteria. Things like “cut this 800-word draft down to 400 words, keeping these three points,” “give me five punchier titles for this article,” “find the three most likely failure points in this plan.” It has a completion state — you can confirm at a glance whether it’s done, so it produces something usable in one pass.

The same thing, framed as a direction versus a task, produces wildly different results. “Help me fix up this article” is a direction — you’ll go back and forth with AI for many rounds and still not be satisfied. “Turn the passive sentences in the second paragraph into active voice, and break up any sentence over 30 words” is a task — it gives you something usable in one shot.

Why Direction Is So Addictive

Understanding the difference is only half of it — you also need to understand why direction is so tempting. Because it satisfies two illusions at once:

First, the illusion of “making progress.” Because there’s no completion state, every round of conversation feels like moving forward — you keep feeling “almost there, a bit more discussion and it’ll be clear.” But progress with no endpoint is, in essence, spinning in place.

Second, the illusion of “growing.” Chatting fluidly with AI about direction feels good — no friction, no need to grind through hard thinking. But it’s exactly that comfort that deceives you. There’s a line that stings- when you have a really enjoyable conversation with AI, it often means you didn’t grow that day, because you didn’t hit any friction. Direction-style rambling is using pleasure to mask the fact that nothing got done. (I explored this separately in Friction Is Growth .)

So “giving direction” isn’t just inefficient — it also tricks you into believing you’re thinking deeply, when you’re actually just enjoying a conversation that produces nothing.

How to Cut Direction Into Tasks

So what’s the right approach? It’s not “don’t have a direction” — it’s that you decide the direction, then cut it into tasks and hand those to AI.

Three concrete steps:

1. Think through the direction and the standard yourself first. What you’re going to write, for whom, what effect you want, what counts as good — this is direction, this is your judgment, and you must have the answer yourself. Don’t ask AI “what do you think” at this step — that’s outsourcing the direction too.
2. Break the direction into concrete tasks, each with a completion state. One big direction gets broken into a string of small tasks, each with clear acceptance criteria. “Write a good article” breaks down into: outline it → draft the introduction (under 150 words) → add a concrete example to each point → replace colloquial words with precise ones → generate three titles.
3. Assign tasks one at a time, and check each one off. Once each task is done, check whether it meets that standard, then move to the next. This way, AI is always doing “work with an endpoint,” and you’re always holding the steering wheel.

You’ll find that once you start doing this, your conversations with AI suddenly get shorter and more concrete — because every round has a clear start and end, instead of an endlessly expanding ramble.

Keep Direction for Yourself, Hand Tasks to AI

Behind this principle is actually a bigger division of labor:

Direction — deciding which way to go, what trade-offs to make, being accountable for the outcome — is your job, and it’s precisely where your growth happens. Task — executing a decided direction efficiently — is AI’s job, and it does this fast and well.

Hold that line, and you get two benefits at once: you get to use AI’s execution power, without outsourcing “thinking and judgment” — the very thing that makes you stronger. Conversely, if you hand direction over to AI too, you’ll end up with an assistant that gets better and better at conversation, and a version of yourself that gets worse and worse at making up its own mind.

So next time you open AI, don’t rush to ask “what should I do about this.” Decide the direction yourself first, then ask yourself: is what I’m about to assign it a task with a completion state, or a direction with no endpoint?

Give it tasks, not directions. This one habit alone will make every collaboration with AI actually land something real.

Related reading: Friction Is Growth | Installing Quality Gates Into Your AI Workflow | Series overview: From Information to Creation .