▶ YouTube Summaries

• YT :: https://www.youtube.com/watch?v=GEkoGCFpY8c
• Original title :: How to stay positive (in the age of AI)

A react to chapter five of Scott's "The true cost of AI coding", where Aaron Francis explains why AI coding doesn't wreck his mental health. Prime's own reading: the loss isn't bad code, it's that the slow, tactile loop of coding-while-designing was the filter that separated good ideas from dross. With that filter gone every idea becomes buildable, which turns into mania rather than productivity. The antidote is narrowing ambition to one or two deep things you have a moat in, and rejecting the manufactured urgency that says a window is closing.

What actually got lost

Prime's grief isn't about AI writing bad code — bad code predates AI. What he misses is sitting down and typing: the flow state where fluency in Go, Zig or Odin meant his fingers were just a translation layer for ideas. He accepts that this is now a hobby activity, like horse riding — still practiced, still occasionally paid, but no longer how the work gets done.

He also pushes back on the framing that AI introduced draining work. Energizing versus depleting tasks existed before. The senior-engineer superpower was always being able to grind through monotony that stalls a junior; most programming was never compiler-from-first-principles, it was pipe A into pipe B.

The design loop that AI collapses

The central argument. Pre-AI, an idea arrived half-formed and implementing even a basic version cost 10x-100x the time spent conceiving it. That implementation time was not overhead — it was where design happened. Shaping the data one way suggests a UI; building the interaction surfaces the real problem. Coding fed design and design fed coding, and the product that fell out had coherence because so much active brain energy had been spent inside it.

AI compresses that time to near zero, and the design time goes with it. What replaces it is myopic: generate, look, decide now, regenerate. Prime's own example is the agent-driven test harness he built for his game — a work queue showing every trial run, personalities, graphs, live state. "Everything I asked for but nothing I actually wanted." Almost nothing on the screen is useful, because he never spent the hours inside the problem that would have told him what to build.

The barrier to entry, in other words, was the refiner. Slow decisions forced you to distinguish 99% dross from the gold. Remove the barrier and every idea has room to grow, which sounds liberating and feels like psychosis — Aaron's word for his own first phase, "almost like infinite possibility."

Infinite possibility is a trap, not a gold rush

Aaron was ripping tokens until midnight or 2am with a wife and four young kids at home. Prime's read: don't undersell that, it will make you feel insane. The pull is constant and granular — waiting for a kid's soccer game, kick off one agent on one tiny UI bug — until you're working seven days a week without noticing.

Aaron's own exit from it wasn't burnout but an economic realization: this is less of a gold rush than it looked, because every other developer has the same godlike power. Building 40 products is pointless when nobody buys anything anymore since they're all building it themselves. Prime adds the anecdote of sitting with seven creators at Open Sauce, three of whom were building their own video editor with Claude Code.

Against the closing-window hype

Prime leans on George Hotz's "love LLM, hate hype": the rhetoric about a closing window, a perpetual underclass, falling hopelessly behind, is mostly designed to make you feel bad about yourself and move to San Francisco. The jump from "fancy autocomplete, smart compiler, better search" to "it owns the whole light cone" is a straw man. Hotz's line that Prime likes best is that the people perpetuating this are terrible people, and the justice is that this is how they feel inside all the time themselves.

He gets emails from people describing it as drowning, being pulled out by riptide. The premise doesn't even cohere: if all the white-collar jobs go, who is left to sell to.

Change the shape of your ambition

Aaron's fix, which Prime endorses: scope down to one or two things and increase vertical ambition on them. Broad to deep, because depth is where a moat is. The payoff is that you can go to bed — you're not going to get scooped on Twitter tomorrow, and even if something superficially similar launches, a thing built slowly until it has genuinely intuitive feel is the thing others have to chase.

Prime admits he feels the pressure anyway. His game is 35k lines, every change needs performance measurement, there's an art backlog, he wants it on a phone, and he wants to play it. He knows the temptation to slop it up and go fast, and knows objectively that the point of the task is to love the thing you're creating — and still finds it easy to bamboozle himself.

Make room for the part you like

The closing practical advice, by analogy with workouts: the ideal routine is worthless if it's so hard you never do it. Find the thing that makes you come back tomorrow. If you like programming, carve out an hour first thing and keep one feature that is just yours, hand-written the way you want it, while prompting for the rest — set up the interface, give me the d.ts, make it like this, because I'm writing that part myself.

This isn't zero-sum; nobody has to become an agent herder building harnesses that build harnesses. Prime suspects the main outcome of most such setups, his own test framework included, is burning a lot of tokens and handing AI providers a lot of money.