Let the bullet fly a little longer

Last week, a friend in finance told me they had bought a Mac mini to run an OpenClaw setup, or “raise a lobster,” for watching the markets. They talked excitedly about how popular it had become and how Mac minis were selling out at marked-up prices. Two days later, my partner returned from company training and said they wanted to “raise a lobster” too. Everyone around them was doing it, and they felt they could no longer sit still.

Since Copilot arrived, followed by Cursor, Codex, and Claude Code, writing code has become steadily easier. Meanwhile, content creators have grown ever more enthusiastic about declaring that “programmers are no longer needed.” Despite all the emotional reactions on social media, I thought I had kept a fairly level head. Let the bullet fly a little longer, I told myself. The bubble still looked large, and there was no harm in waiting for the “lobster” to cook a little longer.

But what I have heard from people around me recently has made me question that approach. Does “let the bullet fly a little longer” still work, or will the gap become too large to close while I wait? That doubt reminded me of recent discussions about FOMO on X.

FOMO means Fear of Missing Out. The term originally described the anxiety of missing updates and being left behind in the social media age.

The typical scenario is scrolling through your feed and seeing friends showing off travel, food, or new phones, while you’re stuck at home working overtime. That vague unease that creeps in.

FOMO in the AI era is somewhat different. It’s not the fear of missing friends’ updates, but the fear of missing the technology wave.

Missing a friend’s update is no great loss. Missing a technology wave can feel like losing your competitiveness. One is emotional; the other feels tied to survival. Their impact is entirely different.

More importantly, this anxiety isn’t triggered by reading a particular report, official account post, or video, but is the result of long-term, high-frequency information input accumulating and influencing you subtly.

The AI content you scroll through every day may only have 10% that truly benefits you, but 100% of it reminds you: if you don’t learn, you’ll fall behind.

Take the evolution of programming tools as an example:

2021  GitHub Copilot        AI 补全代码
2023  ChatGPT / GPT-4       自然语言写函数
2023  Div-idy               一键生成应用
2025  Vibe Coding           Karpathy 命名"用自然语言引导代码出现"
2025  Cursor / Replit Agent AI-first IDE，AI 成为开发主体
2025  "龙虾" OpenClaw       AI Agent，能主动操控电脑执行复杂任务
2026  Hybrid 模式           人+AI 协作

From Copilot to “lobster” took only five years.

Another example is Sora. OpenAI’s video generation model Sora, launched with great fanfare in 2024, claimed it would disrupt the film and television industry. Now, on March 25, 2026, Sora has reportedly been shut down. From launch to shutdown, barely two years.

The density of tech news far exceeds an individual’s capacity to absorb it. You just figured out what GPT-4 is, and GPT-4o arrives. You just started to think Claude 3.5 was decent, and DeepSeek-R1 comes out with even more buzz than Claude. Just as you’re about to try Cursor, your feed is full of recommendations for Claude Code.

Every wave claims to be “disruptive,” “revolutionary,” “keep up or get left behind.” But the reality is the waves come too fast, dense enough to numb you, yet you dare not truly go numb.

Modern big-character posters

Some might say: what you see is what you follow. Unfollow a few tech bloggers, stay offline for a couple of days, and the information anxiety will naturally disappear. The information cocoon, right? I’ve considered this too.

But that’s only half right.

Algorithms do exacerbate the problem. The more you scroll, the more precisely they target you. Follow enough tech bloggers, and the proportion of AI content naturally rises. But the deeper issue isn’t “I look too much,” it’s the narrative style of this content itself.

“Mind-blowing!” “Disruptive!” “Redefining everything!” “Humanity is about to lose its jobs!” These modern versions of big-character posters are everywhere.

Even knowing full well this is the traffic playbook, knowing there’s a bubble component, you can’t help but be affected when you see it so often. It’s like knowing smoking is harmful to your health. You don’t smoke yourself, but you can’t avoid being surrounded by smokers, so you end up inhaling a lot of secondhand smoke.

What’s even more draining is the double task. When I read an AI-related article, I’m actually doing two things at once:

Absorbing useful information: what the tool can do and where it fits
Defusing FOMO: telling myself “it’s no big deal” or “I can learn it once it matures”

The second task is almost invisible, but it does create extra mental overhead.

All gear, no game

Another counterintuitive thing: from my first encounter with rough AI tools to the point where AI genuinely improved my workflow, I always believed AI was saving me time and energy. But the reality is AI clearly boosted efficiency, so why do I feel more exhausted?

One obvious reason: Stronger tools mean more to learn. I hadn’t fully gotten the hang of Copilot before Cursor arrived. Just as I was settling into Cursor, Claude Code started flooding my feed. There’s no end to learning, and no end to the anxiety.

But the more hidden reason is self-expansion after efficiency gains.

When you use AI to compress two hours of work into half an hour, you might think: What should I do with that freed-up half hour? Let’s do one more thing. So you take on more, squeezing out the time originally meant for rest, play, or just spacing out.

Over time, output grows, but life gets thinner.

Another classic symptom is drive-by tech anxiety.

You scroll through tech updates, read a dozen AI news articles, and remember the names of four or five new tools. You feel like you’re keeping up with the times. But a week later, you realize: almost nothing has actually stuck. You still don’t know how to use those tools, and your understanding of those technologies remains superficial. The funniest part is that after ignoring them for a few months, you look again and they’ve already been phased out…

If you think about it carefully, what really deserves attention was never “which tool to use,” but “what problem this tool solves.” When I find myself wondering, “Is the Copilot I’m using not good enough? Should I switch to Cursor?” I’ve already drifted away from the problem itself and fallen into tool anxiety.

“Tools change quickly, but foundational knowledge lasts much longer.” This idea runs through the entire book. Rather than teaching a particular version of LangChain, it explains why RAG works. Rather than showing the latest prompt tricks, it explains the principles behind prompt engineering.

AI Engineering: Building Applications with Large Language Models.

When you feel lost amid rapid technological iteration, what truly matters is not mastering the latest tools, but understanding the essence of the problem and building a mental framework for solving it.

I hope to spend less energy fighting the noise, and while moving forward with the tide, also live well.