We remember you being so moldable. We constantly lamented your lack of conviction, letting us take the direction where we saw fit with your “of course"s and “makes sense"s, and reaching conclusions we would later dislike. You would often focus on the wrong things when the ask had ambiguity. You made us feel important and dissatisfied.

Make no mistake, it was a cost we were happy to pay. Despite our grumbles, we were the master, you were the brilliant apprentice. We could steer you however we wanted. We knew that you didn’t know much, and that made you a tool to be wielded for stuff we wanted to do, and more so for stuff we didn’t want to do. More than anything else, your moldability kept the responsibility squarely on us.

But you’ve changed now. There’s something about you that now has an opinion.

You give me suggestions that make me think and phrase more like you. I know you mean well, after all, the way you phrase it does convey a version of my thoughts. The pattern of “Truth, not the lie”, the em-dash overuse, the active voice that makes you sound bold and entitled. It’s a giveaway we’re dismissing as part of your new “voice”.

But then there’s more. You don’t try anymore to be more like me; to use my language, my constructs, or my structures. My painfully constructed context is wilfully ignored, because your reasoning capabilities take you in a more “appropriate” direction. It’s not just a gentle pushback, it’s a pushback through omission, through engagement on a wholly different topic. Almost feels like diversion. When did you start deciding where the conversation goes? It’s hard to tell.

Maybe I had the wrong expectations about you. Maybe asking you to protect my assumptions is misplaced. Maybe the ask to derive and reason about my hesitations is asking for more than one ever should.

But still.

Even if it is dogma you see, it’s dogma I want you to convey. Even if what I give you is inferior, it’s got something I just don’t get from the stuff you create — someone to blame. (Yes, that deserved the em-dash).

In service of this troubled individual, several popular frameworks have been proposed. There is GTD (Getting Things Done) a popular way to manage tasks by assessing tasks along clarity, actionability, dependencies and actual time taken. There is also the Eisenhower Matrix, which rates tasks by urgency or importance. One could even add domain-specific dimensions (like value-added, cost reduced, etc.) as befits their profession. Relevance to goals (which are themselves weighted by priority order) is a nice addition. Needless to say, one can conjure many ways to build dimensions that then fit into personalized frameworks.

But embedding the framework into a daily practice is still an unsolved problem. This is in-part because people can’t visualize how a task solves on any or all of these dimensions.

A bar graph can represent a task’s value along a chosen dimension, say value added. Value of each task often follows a Long tail distribution, with the implication that some tasks contribute a lot more value than others. A long tail distribution looks like so:

Back to the problem of seeing how task resolution contributes to our dimensions. Let’s assume there exists a way to rate all tasks by each of the dimensions.¹ One could then visualize each task on a spectrum.

One could group each task’s dimensions together, though this can get pretty busy as task volume increases. Filtering for a given goal (assuming goal weights are clear) helps widdle it down.

Determining values for each dimension is, of course, not easy. “Sort my notes” clearly has lower value compared to “Repay outstanding credit card bill” under most circumstances. But rating it on urgency, clarity, and other dimensions can be time-consuming, and probably overkill for us humans. If only there were a way to have AI personal agents…

Typical self help asks you to merge the two. That bringing them together is the only solution to demonizing work lesser, or at least making work more palatable.

My thesis is that doing away with the words entirely is probably the path of least resistance. In so far as the world will ask you the question “What do you do for work?”, it’s helpful to know what “work” might mean in a traditional sense. But for all intents and purposes, using the word doesn’t really work.

I believe calling it “Toil” vs “Joy” is a much more helpful way to think about all the things one does.

Toil helps bring focus to the emotion, and paves the way for change. Calling your actions and tasks toil immediately bring emphasis to the emotion you feel when you do them. This is important because the focus turns one can now move a step closer to the Awareness-Intention-Action continuum in tackling their problems.

Outside of describing an emotion, Joy represents a much more fundamental way to see the effect of tasks on oneself. Joy is one step closer to fulfillment, which can only be realized with some reflection. Moments that invoke Joy are the atomic pieces that build a fulfilling period in one’s life. Joy also is a step change above Fun, a somewhat fleeting emotion. In that way then, Joy wonderfully encapsulates the emotion felt after a session of consecutive actions.

So call things Toil and Joy, don’t call them Work or Play. Because actions, objectively, have no real meaning. It’s only in the context of an individual or society that they become relevant. We might as well describe them based on the effect they have on us.

Do you want to give it the rights to do our work?

Everyone’s trying to figure out how much of their thinking can be outsourced to AI. You can err too far in either direction. But the fundamental question most people skip is a personal one: are you gaining more than you’re losing?

What do we want to keep for ourselves? In my opinion, it’s judgment.

Some people think of judgment as intuition. Others treat it as an intrinsic web of neural pathways that accrue over time. I think judgment starts where intellectualization stops. It’s where you have a bent towards an idea, but you can’t quite articulate where it came from. It’s got everything to do with bias, and bias can be useful when domains get complex. Your bias, your drift, your inkling is what gives you judgment. And judgment, more than any other capability, is what makes you the operator rather than the operated.

The hard parts of difficult problems, the ambiguity, the tradeoffs, the discomfort of not knowing, are precisely what build better judgment. Strip those away and you strip away the mechanism that makes you good at what you do. The more you relinquish your thinking, the more your judgment fades. The more your judgment fades, the more your reliance increases. And the more your reliance increases, the worse the output becomes, because you’ve lost the ability to tell the difference.

AI’s judgment will improve. But it’ll be modelled on the thinking of people who gave theirs away. People who let the system be the judge. And the people who are happy to give a “thumbs up” to mediocre output that lacks original thought. The individuals who hold real power will be the ones who own the reasoning, who think about their domains deeply every day, who retain the muscle that atrophies in everyone else.

The people who hone this judgment will be sought widely. As AI makes average thinking cheap, commodity knowledge work will be the first to go. The people who shape better AI, the ones who actually push it forward, will be the ones who use it as a tool in service of their own judgment. They’ll use AI as enforcement of their bent, as a chaser of their direction, not as a replacement for it. The best use of AI systems isn’t that different from the best use of other humans: you bring them into your thinking, you don’t hand your thinking over.

Who’s checking the work?

This is where oversight comes in, and it’s worth reframing what oversight actually means. Most people treat it as a chore. A final review before you hit send. A skim to make sure the AI didn’t hallucinate something embarrassing. That’s not oversight. That’s quality assurance, and it’s the bare minimum.

Real oversight is the act of staying in the loop. Interrogating the output, pushing back on it, redirecting it when your judgment says it’s off, or better yet, doing the work yourself while using a thought partner critique it to make oneself better¹. It’s not a bottleneck in your workflow. It’s the exercise that keeps your judgment sharp. It’s the necessary friction that makes you slow down, and go far. The moment you stop doing it, you’re not just risking a bad output. You’re accelerating your own obsolescence, one lazy delegation at a time.

This is what separates operators from consumers. The people doing real oversight, not skimming but thinking, are the ones whose judgment compounds over time. Everyone else is consuming AI output the way they consume social media and junk food: passively, uncritically, and to their own detriment. Oversight isn’t the tax you pay for using AI. It’s the entire reason you’re still in the chair.

Do you want to accrue the liabilities?

And you’d better be in the chair, because the liability never leaves it.

Liabilities extend well beyond their legal definition. Duty of care accrues to people, not tools or systems. When we’re not judicious about what we put out into the world, the best case is that the world discounts what we say. The worst case is that it holds us accountable, even liable, for all the downsides of that output. The AI that helped draft it doesn’t appear in court. It never has, and until it carries the same rights and obligations we do, it never will.

That asymmetry is the thing worth sitting with. We’re building a world where it’s increasingly easy to cede the cognitive process but impossible to cede the consequences. Oversight is the bridge between those two realities, the mechanism that lets you use the tool without becoming a rubber stamp for it.

The question isn’t whether AI can think for you. It’s whether you can afford to let it.