Four Questions I Ask Before I Use AI for Anything

Most people pick up an AI tool and ask: what can this thing do?

That’s the wrong starting point.

The better question is: what is actually worth doing? And of that, what’s worth doing myself?

I’ve used AI in my work long enough to have built a few mental frameworks that tell me when and how to reach for it. Not out of enthusiasm, and not because it’s new. Because it genuinely cuts the time, effort, and stress of getting work done, and I’ve gotten more careful about why I’m using it in any given situation.

Four frameworks do most of the heavy lifting. Together they tell me whether to act, what to act on, which parts of a process deserve AI and which deserve me, and how to run the whole thing without burning energy on work that shouldn’t exist. If you’re trying to figure out where AI actually fits in your work, not theoretically but practically, on a Tuesday when you’re already behind, this is where I’d start.

Framework 1: The Three Circles

Before you think about AI at all, there’s a prior question: is this thing worth spending mental energy on?

I’ve carried this model for a long time. It comes from martial arts, where the three circles represent ranges of defense. The inner circle is your physical body. The middle circle is the reach of your extended fist. The outer circle is the reach of your extended leg. What stuck with me was how cleanly it maps to the serenity prayer: the wisdom to accept the things you cannot change, to change the things you can, and to know the difference.

Applied to work, it goes like this. The inner circle is everything you can fully control. Your own mental state, how you react, the decisions that are yours to make. Own them. The middle circle is everything you can influence. Your words, your actions, your effort. You don’t control outcomes here, but you have real pull, so work the influence. The outer circle is everything else. Things you can be aware of, can prepare for, can try to reduce, but cannot change. Don’t spend bandwidth worrying about those.

There’s one thing you can do with the outer circle: automate the monitoring of it. You can’t control what happens out there, but you can set up systems that watch for it and trigger a response when something moves. A competitor makes a change. A market signal shifts. A metric crosses a threshold. You didn’t cause it and you can’t stop it, but you can make sure you know about it fast enough to respond inside your circles of control and influence. That’s the right use of AI on outer-circle problems. Not fixing them. Watching them.

The reason this framework comes first is that it’s a triage tool. Before you build a workflow, before you automate anything, ask which circle a task lives in. If the answer is the outer one, the work isn’t automation. It’s monitoring, so you’re ready to act when the circle shifts.

There’s a practical side too. I came home one night mentally done. The house was a mess and I didn’t have the executive capacity to figure out where to start. I opened Gemini, took a photo of the room, and asked it one question: what’s the simplest thing I can do in the next 15 minutes that would have the biggest visible impact with the lowest effort? It gave me three things. I did them. Took another photo. Repeated the prompt. In 45 minutes the room was sorted. I didn’t mind doing the tasks. I just didn’t have the bandwidth to decide which tasks. That’s a middle-circle problem. I could influence the state of the room. I just needed help figuring out where to push first.

Framework 2: Eliminate, Then Automate, Then Delegate

Once you’ve identified what’s worth working on, this framework tells you how to approach it. I’ve been running it since about 1998. Get good at a job, get efficient, then ask three questions in sequence: what can I eliminate, what can I automate, what can I delegate?

The sequence is the whole point, and most people get it wrong. They look at a process and immediately ask how to automate it. But if you haven’t eliminated first, you’re building machinery around something that shouldn’t exist. You’re making a broken thing faster.

Eliminate first. Look at each step and ask honestly: does this need to happen at all? Does it need to happen this way? Is it creating value, or just the appearance of activity? A lot of what fills workdays is in that category: reports nobody reads, meetings nobody needs, formatting that serves no audience. Get rid of it before you touch anything else. Then automate. Once you’ve kept only the steps that earn their place, ask which can run without you. Repeatable, definable steps with clear inputs and predictable outputs are the candidates. Give AI the process, the standards, the constraints, and let it run. Then delegate. What’s left still needs doing, but not necessarily by you. Delegate the execution. Keep the judgment.

There’s no point delegating something you should have eliminated. There’s no point automating something you haven’t decided is worth keeping.

I built this habit before AI tools existed. What’s changed is the scope of what’s now automatable. Research, drafting, formatting, scheduling, summarizing: all of it can run on an LLM with the right setup. The framework is the same. The surface area it covers has expanded.

Framework 3: The $9 Task vs. The $100 Task

Not everything you do at work is worth the same amount, and most people have never stopped to look at which is which.

Say you’re making $60,000 a year. That’s roughly $30 an hour. But not everything you do in a day is a $30-an-hour task. Some of it is $100-an-hour work: judgment, discernment, taste, creativity, the expertise you’ve built over years, the calls only you can make. Some of it is $9-an-hour work: taking information from here, shaping it a little, putting it there. Data entry. Formatting. Scheduling. Updating the project board. Sending a note that a thing is done.

At least half, if not an easy majority, of what knowledge workers do every day is $9-an-hour work. That’s not an insult. It’s how most jobs are structured. The $9 tasks are the scaffolding around the real work. The problem is that scaffolding takes up most of the day. By the time you’ve processed email, updated statuses, moved data, and handled the communication overhead, the hours you had for actual $100 work are mostly gone.

I’ve been a creative professional for going on 30 years. Ideas have always been weighed against the cost of execution. We’re entering a period where the cost of execution is approaching zero, which means how well you can make a thing takes a back seat to how bold your idea is and how sound your strategy is. The $9/$100 read is how you sort tasks before you hand anything off. Once you start seeing it, you see it everywhere.

Framework 4: The Four Quadrant Matrix

The Matrix is the most detailed of the four. It gives you a precise read on any process, task by task, and tells you exactly what role AI should play in each part. It’s not a task sorter. It’s a process dissector.

Don’t start with a list of tasks. Start with a single job responsibility you’re accountable for end to end. The one where, if someone told you tomorrow it was permanently off your plate, you’d feel immediate physical relief. Then crack it open and write down every task inside it, in sequence, as granular as you can go. That list is what you run through the matrix.

The matrix has two axes. Skill: things you’re good at on one end, things you’re not on the other. Enjoyment: things you like on one side, things you don’t on the other. That gives you four quadrants. Things you’re good at and like are your highest-value work, the stuff that defines your professional identity; AI’s job here is to clear everything else so you have more room for it. Things you’re good at but don’t like are skilled work that’s become a drain; teach AI to do them your way, to your standards, and your skill stays in the output while your attention stops being the cost. Things you’re not good at but like are on their way to becoming your highest-value work; AI can coach you while you build the skill. Things you’re not good at and don’t like are your most expensive $9 work; hand them off entirely.

One thing the matrix surfaced for me that I didn’t expect: it revealed not just what tasks I liked, but what I actually liked about them. I used to think I loved doing research. What I realized when the deep-research tools got capable was that what I loved was the discovery. The learning. The moment something new clicks. The LLM could do two weeks of reading in five minutes, and I still got the hit I was after, just faster.

Sometimes what you love is the output, not the process. The process is just friction you’ve gotten used to calling work.

Where to start

Take the one process in your job that you dread most. Map every step from beginning to end. Be specific. Run each step through the Three Circles: can you control or influence it? If it lives outside both circles, set up monitoring if it matters, then let it go. For everything inside your circles, run eliminate, automate, delegate, in that order. Then look at what’s left and ask whether it’s $9 work or $100 work. Then plot the remaining tasks on the matrix, skill against enjoyment, and find what’s been costing you most.

Take what you find to an AI you trust. Tell it the process, the outcome you want, the context it needs, the constraints that matter, and ask it to help you build a workflow that runs the whole thing with you stepping in only where your judgment changes something. Set a target of 90 percent time reduction. It sounds aggressive. For most processes, it’s accurate.

That’s the work. Not a life hack, and not a shortcut. Just a cleaner way of deciding what gets your time, and making sure the answer is the work that’s actually worth it.