How to Think in an Algorithm's World

I caught myself again.

Scrolling LinkedIn, late, the feed thick with AI content engineered to make me react. One post: "If your job involves a spreadsheet, you have 18 months. The people telling you otherwise are lying to keep you calm." My thumb moved toward the comment box — I had seen the adoption curves up close, I knew it was not that fast, and I wanted to say so.

Three posts down, the opposite: "The 'AI will take your job' panic is just fear-mongering from people who have never shipped anything. Adapt or stay scared." And I felt the pull again — this time to agree, to repost, because it flattered the part of me that has watched the doomer timeline fail to arrive.

Two posts. Contradicting each other. Each one pulling me toward a reaction. And here is the part that unsettled me: neither one matched what I actually believe. My real view is more nuanced than either, and quieter, and it would never survive contact with that feed.

I did not comment. I did not repost.

But I sat with how close I came — twice, in the span of one scroll.

That is what I want to write about. Not AI specifically, though I will get there. The deeper problem is this: we are living inside systems optimized to pull us into camps, and most of us do not notice the pull because the pull feels like thinking.

The mechanism is not a conspiracy. It is incentives.

Recommendation algorithms optimize for engagement. Engagement tracks emotional activation. The cheapest emotional activation is tribal: us versus them, right versus wrong, based versus cringe. So the feed you see, on almost any platform, is structurally biased toward content that makes you feel like you are on a team.

Once you feel like you are on a team, you start reaching for conclusions your team would approve of before you have done the work of actually thinking.

This is happening to everyone.

Including me. Including you.

The question is not whether you have been captured. The question is where, how deeply, and what you are doing about it.

The Frame: Practice, Posture, Conclusion

I have found one frame useful: the practice, the posture, and the conclusion.

The practice is the discipline of genuinely understanding opposing views before judging them. Not the weakest version. Not the cartoon version your side likes to mock. The strongest version each camp would recognize as its own.

This is not a trait I possess. It is work I do. Sometimes well. Sometimes badly.

When I am tired, rushed, or emotionally activated, I do it worse. When I am rested and the stakes feel lower, I do it better. Anyone telling you they naturally see all sides is probably selling you something.

The posture is the stance of not being on a team too early.

I hold this lightly. It is tactical, not identity. On most questions I can find a reasonable answer from multiple starting points, and declaring allegiance before doing the work is premature.

But "not on a team" is not the same as "no conclusions."

It means: no conclusions yet.

The conclusion is where you actually land after doing the work.

This is the part that gets missed: neutrality is not the goal. Neutrality is the starting point. The goal is to reach conclusions you have earned — conclusions that came from the practice, not from the tribe.

Sometimes those conclusions will put you on a side. That is fine. A side you reasoned your way onto is different from a side the algorithm sorted you into, even if they happen to look the same from the outside.

There are two failure modes.

One is picking a team before thinking.

The other is never reaching a conclusion at all — hiding behind "I see all sides" to avoid the cost of having a position.

Both are failures.

The practice is the narrow path between them.

AI as the Test Case

AI is where this frame has mattered most for me.

I have spent time in most of the camps.

I have been doomer-adjacent: genuinely worried about capability overhang, job displacement, and the failure modes of systems we do not fully understand.

I have been accelerationist-adjacent: convinced that slowing down hands the future to worse actors, that the upside is civilizationally significant, and that fear can become a failure of imagination.

I have held versions of the ethics-and-labor critique.

I have not spent much time in the "it is just autocomplete" camp. I will be honest: those people worry me, because dismissal is its own form of not-thinking.

Two years ago I leaned harder toward the doomer end than I do now. I believed, with real conviction, that the transition might happen fast enough to cause major societal chaos — mass displacement on a timeline too fast for institutions to absorb.

I read the doomer essays. I nodded along. I felt like I was being clear-eyed about risk while everyone else was in denial.

Then I started doing the work — not abstract thinking work, but deployment work.

I was building a financial reporting system for a multi-entity organization that had grown through acquisition. Five different source systems, five different ways of tracking revenue, five different definitions of what counted as a "sale." I used AI heavily to do the technical work — write the SQL, structure the data model, generate the DAX measures, build the daily sales-versus-budget report. The technical execution was fast. Genuinely impressive.

What was not fast was getting the numbers right.

I spent months in meetings with the VP of Finance, going through variances line by line. This number does not look right. That subsidiary recognizes revenue differently. We stopped counting that program in the totals after the second acquisition closed. That category had a one-time adjustment in Q3 that you would not see in the data.

Every one of those was a judgment call that lived in her head. Not in any system. Not in any document. Built from years of watching the business and noticing where the seams were.

AI wrote the pipeline. AI wrote the report. AI could not have those conversations, because AI did not know which numbers were supposed to feel right.

And that, multiplied across every business deploying these tools, is the gap.

Call it the capability-adoption gap.

The doomer model often assumes capability translates into impact on roughly the same timeline. It does not. Organizations are made of humans, incentives, legacy systems, budget cycles, and — most importantly — accumulated judgment that lives in specific people who have been watching specific things for a long time.

That friction is not always wise. It is not always efficient.

But it is real.

And in this case, it acts as a shock absorber.

That did not make me conclude AI is overhyped. AI is very real. The upside is very real. The disruption is very real.

But "AI will move fast enough to break everything before we can adapt" stopped matching what I was seeing with my own eyes.

The accelerationist camp has its own blind spot.

The physical constraints are real: chips, power, rare earths, water, cooling, grid capacity, industrial supply chains. We are not scaling intelligence in a vacuum. We are scaling it through the physical world.

The doomers often underweight this because it complicates the "unstoppable superintelligence" narrative. The accelerationists often underweight it because it complicates the "scale solves everything" narrative. The ethics camp is focused, reasonably, on present harms. The dismissal camp does not think it matters.

So one of the most concrete near-term constraints on the entire AI trajectory is also one of the least emotionally satisfying to talk about.

It does not belong cleanly to any tribe.

That is usually a sign you should pay attention.

There is one more thing I sit with.

Every prior time humanity went through a shift this large, we could not clearly see the other side of it. We could not imagine the jobs, institutions, tools, or forms of meaning that would emerge.

People trying to forecast industrial society from inside agrarian society were wrong about almost every specific. People trying to forecast the internet economy from the early web were wrong about most specifics too.

The current doomer case often rests on: this time, we can see the end state, and it is jobless.

The current accelerationist case often rests on: this time, we can see the end state, and it is abundance.

Both are claiming visibility through a fog that has historically been opaque.

The humbler position is that we do not know. And every prior time we did not know, humans found something.

So where do I land?

Roughly here:

The AI transition will likely be slower than the doomers think and more constrained than the accelerationists think. It will still cause real displacement.

My base case is that central banks will print to backstop the reskilling transition — not as a prescription, but as a forecast. The system is structurally incapable of absorbing that kind of unemployment any other way.

That is part of why I own Bitcoin. Bitcoin was made for this AI transition.

The optimistic case — and I hold it genuinely — is that AI gives humanity leverage to build a more abundant future for more people than the current system serves.

The cautious case — which I also hold — is that how we build matters enormously, and careful stewardship is not the same thing as slowing down.

That position agrees partially with multiple camps and fully with none.

I did not start there.

I got there by doing the practice, badly at first, better over time, and being willing to update when what I was seeing stopped matching what I was saying.

Cognitive Sovereignty

This is the thread that ties back to everything I write about.

Sovereignty is usually discussed in terms of money, health, time, or ownership.

But the prerequisite for every other kind of sovereignty is cognitive sovereignty: the ability to reach your own conclusions.

You cannot own your money, your body, your calendar, or your future if you do not own your thinking.

And right now, the systems we spend much of our waking life inside are designed to move your thinking toward conclusions that serve their engagement metrics, not your life.

Attention capture is sovereignty failure.

It is the first domino.

Once your attention is captured, your framing gets captured. Once your framing is captured, your conclusions follow. By the time you are defending a conclusion, you may have already forgotten you did not arrive at it yourself.

The practice is how you push back.

Not by refusing to have views. That is another kind of capture, dressed up as sophistication.

You push back by doing the work to make sure the views you hold are views you would still hold if you had not been served the inputs the algorithm chose for you.

I do not always succeed at this. Nobody does. The feed is stronger than any individual, most of the time.

But I have come to believe the practice matters more than almost anything else I do, because everything else is downstream of the conclusions I reach.

And the conclusions I reach are downstream of whether I am doing my own thinking or borrowing someone else's.

Sovereignty is not just about what you own.

It is about whose conclusions you are reaching.