A few months ago a founder asked me what I meant by “philosophical thinking.” He’d sat through a graduate philosophy course in Russia some years back and was, to put it gently, unimpressed. His exact phrase, in translation, was “professors shitting on engineers, physicists, and mathematicians.” He couldn’t reconcile that with anything resembling useful thought.

Fair. I wouldn’t reconcile it either.

But the answer to “what’s philosophical thinking” doesn’t live in graduate seminars. It lives in three writers most operators never read - Simone Weil, Hannah Arendt, and Carol Dweck - and the operating system they accidentally built between them.

They didn’t know each other. Their lives don’t overlap in any tidy way. Weil died in 1943 of self-imposed starvation. Arendt lived through the Eichmann trial and gave us “banality of evil” - a phrase that everyone misuses. Dweck is a Stanford psychologist whose “growth mindset” got flattened into a Pinterest quote. Three different centuries of thought, three different countries, three different disciplines.

And if you read them together, they slot into the same architecture: sensors, compute, feedback loop. The same architecture you’d use to design a reliable system at scale. Except this one runs in your head, and most people are missing one of the three layers without noticing.

The Underused Toolkit

A note on why the toolkit is underused. Modern philosophy and psychology lean heavily on women writers - Weil, Arendt, Dweck, Anscombe, Murdoch, Nussbaum, O’Neill - whose work is operationally precise. Most of them never made it into the canon that founders and engineering leaders quote at each other.

There are reasons. Some historical, some about how popular culture filters who gets remembered as “useful” versus “inspiring.” Weil gets shelved in mysticism. Arendt gets reduced to one phrase about Nazis. Dweck gets quoted in motivational decks by people who never opened the book.

The result: a generation of operators who have read every Stoic but can’t tell you what attention is, what thinking is, or what failure is for. The toolkit is sitting there. Nobody reaches for it.

I’m going to spend three more posts after this one going deep into each of the three. Today is the map.

Sensors: Simone Weil

Weil’s word is attention. It looks like a soft word. It isn’t.

For Weil, attention is not focus. It’s not concentration. It’s the discipline of seeing what’s actually in front of you, without bending it into the shape you wish it had. “Attention is the rarest and purest form of generosity,” she wrote. The generosity is in not protecting yourself from the truth.

Her sharper line: “Attachment is the great fabricator of illusions; reality can be obtained only by someone who is detached.” The thing you’re attached to - your version of the product, your read of the market, your reading of the team - is exactly the thing that fabricates the lie you’ll later trip over.

This is the sensor layer. Before you can think about a situation, you have to see it without the comfort filter. Weil makes you do the unpleasant work of taking the filter off.

In an AI-saturated workflow, this layer is now load-bearing in a way it wasn’t ten years ago. A language model will always give you a fluent answer. Fluency is the new comfortable lie. Without Weil-style attention, you read the answer and trust it because it sounds right. The whole rest of your reasoning then runs on hallucination as input.

Compute: Hannah Arendt

Arendt’s word is thinking. She means something specific by it, and it’s not what most people think.

Everyone knows the phrase “banality of evil.” Almost no one uses it the way she meant. She watched Eichmann at trial in Jerusalem and was struck not by his monstrousness but by the absence of one specific quality. He couldn’t think. Not “wasn’t smart enough” - he was perfectly competent. He couldn’t carry on the inner dialogue Arendt called the “two-in-one,” the conversation between you and yourself that produces conscience as a by-product.

“Evil comes from a failure to think,” she wrote. It comes from the substitution of execution for reflection. The man who follows the procedure, who repeats the slogan, who never asks himself “what am I actually doing right now and why” - that’s the failure mode she was naming.

Take it out of the Holocaust context, where it gets uncomfortable to apply. The same machinery shows up at every scale. Decision-by-default in a quarterly review. “It’s how we do things.” “It’s the process.” Boards that nod at the deck because they’re attached to the founder. Engineering teams that adopt the architecture the AI suggested without asking why those tradeoffs and not others.

This is the compute layer. Sensors give you data. Compute is what you do with it - and most of the compute most of us run is autopilot.

AI amplifies this one too, in a different way. It makes thoughtless decisions sound thoughtful. The slide deck has structure. The summary has bullet points. The architecture diagram has labels. Everything looks like the output of thinking. None of it required the inner dialogue.

Feedback Loop: Carol Dweck

Dweck’s word is mindset. This is the one you have to actively rescue from how it’s been popularized.

The Pinterest version is “believe in yourself.” The actual research is sharper. Dweck’s distinction between fixed and growth mindset isn’t about optimism. It’s about what you do with failure as a data point. Fixed mindset reads failure as a statement about your identity (“I’m not the kind of person who can do this”). Growth mindset reads failure as a problem (“there’s something here to face, deal with, and learn from”).

The mechanism is a feedback loop. Failure → data → updated model → next attempt. Identity-coded failure breaks the loop because you stop processing it as data and start processing it as a verdict. Once it’s a verdict, there’s nothing to learn.

Worth noting: there’s a recent critique - Yadav published a paper in 2024 calling out “false growth mindset,” where the concept itself becomes an identity (“I’m a growth-mindset person”). The trap eats its own tail. Real growth mindset includes acknowledging when you’re in the fixed-mindset version of yourself. That’s the honesty layer.

This is the feedback loop. Sensors fed you data. Compute processed it. Failure - which is to say, the discrepancy between what you predicted and what happened - is the signal that lets you update. If you don’t process failure as data, you stop learning, and the next decision runs on the same flawed model as the last one.

AI compresses iteration cycles to the point where this layer matters more than ever. A refactor takes minutes. A test rerun takes seconds. The opportunities to learn from failure happen so fast that without an explicit Dweck-style protocol, you skip past them and never extract the lesson.

The Architecture

Read together: sensors, compute, feedback loop. Take out any one and the system breaks.

  • No sensors: you compute carefully over hallucinations.
  • No compute: you see clearly and then default to autopilot.
  • No feedback loop: you see, you think, and you repeat the same mistake on the next decision.

I keep finding myself missing one of the three at any given moment. Usually compute - I see the situation, I have a clean read, and then I act on the default move because the meeting is in five minutes. Sometimes sensors - I’m so attached to the decision I’ve already half-made that I read the data through it. Feedback loop fails on me less often, but only because I’ve worked on it the longest, not because I’m done.

The next three posts go into each of these one at a time. Weil first - sensors, and the cost of seeing what’s actually there.