Blogs

93 Percent People approved Claude Code’s permission prompts roughly 93% of the time. Anthropic shared that number to make a point: the more prompts you see, the less you read them, and eventually the human check stops being much of a check. Anthropic describes two fixes. One is to let another system make the routine calls: auto mode, where a classifier decides which actions are safe enough to approve. The other is to limit what the agent can do in the first place, so some actions aren’t possible at all. Either way, the idea is to stop relying on someone paying attention to every single action. ...

The Idea There’s an idea going around tech right now. Raj Nandan Sharma puts it bluntly: when AI makes building easy for everyone, good taste is the only real advantage left. The Last Software Engineer says something similar: once writing the code is cheap, the hard part is deciding what should be built. Taste, judgment, call it what you like. I mean the calls that aren’t written in any instruction. What’s worth doing and what isn’t. What to leave out. What actually matters in this situation. How to say something to this particular person. ...

The Argument I got into it with family a while back about protein. My position was that Indian diets are protein-deficient. I think this traces back to how much less meat and fish people eat now compared to a few generations back. Their position was different. They said Indians have thrived on a carb-heavy diet for millennia, so the sudden alarm about protein doesn’t make sense. I pushed back. It’s a different population under different conditions, I said. The loss of animal protein over time is the real gap. We went around a few more times. At some point it started going in a loop. I had said what I had to say. I didn’t agree with anything they were saying. Continuing wasn’t going to produce anything. So I said something like “fine, whatever,” and let it go. ...

A while back I read an MIT study on how our brains process code. The finding surprised me: it doesn’t activate the language centers the way reading normally does. Instead it lights up something called the multiple demand network (frontal and parietal lobes), the same system used for solving complex problem. The researchers said reading code is “not the same as language, and not the same as math.” It’s its own thing. I read that and thought, oh! that’s also how I process everything else! ...

In 2023, I had too much on my plate. To be honest, that’s been true for most of my adult life, but 2023 was when I actually sat with it. Work projects, finances, psychology research (don’t ask), exercise routines, new technology, new shows. I had built a careful system around all of it. A routine that felt like it was moving me somewhere. Then I had to travel. A week away from home, somewhere I didn’t particularly want to be. I couldn’t do even a fraction of what I normally did. By the first afternoon, a feeling I recognized immediately had settled in: guilt. Not the guilt of having done something wrong. The other kind. Productivity guilt. The sense that by stepping away, I was somehow betraying my own ambitions. ...

A while back, a friend and I were going back and forth about a decision at work. I wanted to make a point about how a certain kind of leadership style plays out, so I reached for Elon Musk and Twitter as my example. My friend brushed it off. Not interested, moved on. I sat with that for a second, then tried again. Same point, same structure, but this time using our own company and our own manager. My friend engaged. We had the conversation. ...

At some point I noticed I couldn’t tell which video I actually wanted to watch. Not in a “too many options” way. More like: I’d be scrolling through Instagram or YouTube, something would catch my eye, I’d watch it, and by the end I couldn’t tell you if it had been worth my time. Everything registered as roughly the same signal. I was consuming a lot and feeling like damn, I’ve just spent too much time on “entertainment” again. ...