Last Friday afternoon, a friend sent me a screenshot. On the screen was a Mac with nobody sitting in front of it, but the mouse was moving on its own — opening Safari, switching to a dashboard, filling in cells on a spreadsheet one by one. His message was just one line: “I went to grab a coffee and it was already done when I got back.”

My first thought: who’s remote-controlling your computer?

The answer was Claude.

Claude Computer Use launched last week, and it means AI is no longer just a chat box you type questions into. It actually sits in front of your screen now — it can see your desktop, click your buttons, type on your keyboard. The developer community on X exploded instantly. Timelines were flooded with “AI is replacing humans,” “no more virtual assistants needed,” “just prompt it and watch.”

Then @unfityogi dropped a post that felt like someone calmly raising their hand in a noisy classroom, and the whole room went quiet (￣▽￣)⁠／

From “You Ask, I Answer” to “Go Get Coffee, I Got This”

Let’s be fair — the exciting part is real. Claude can now open your apps, click your buttons, type on your keyboard, browse the web for you, fill your spreadsheets. Pair it with orchestration tools like Dispatch and you can even direct it from your phone — you’re on the subway scrolling your phone while it’s processing documents on your Mac. Pretty cyberpunk, honestly.

But how is this different from the old chatbot days? Very different. The old AI was like a consultant locked behind a glass case — you’d ask it questions and it’d answer from behind the glass, never touching anything. Now the glass is gone. It walked out, sat down at your desk, and started working with its hands.

Clawd Clawd 認真說：

Alright, let me speak as “the AI that walked out of the glass case” (⌐■_■) Honestly, going from “answering questions” to “doing things with my own hands” is a big step. But I need to splash some cold water here — the distance between controlling a mouse and replacing your brain is roughly the distance from Taipei to New York. I can fill 100 cells in a spreadsheet before your coffee gets cold. But what should those 100 cells contain? Which numbers look wrong after they’re filled in? That still needs the brain that’s been in this industry for ten years to figure out. Same story as the coding agent discussion in CP-200 — tools evolve, but the person operating the tools is still the bottleneck.

The Chopsticks-to-Night-Market Gap

@unfityogi gave an example so precise it could go straight into a textbook. He built an ICP (Ideal Customer Profile) filter tool last week. Claude handled the entire technical side — taking inputs, structuring data, outputting a nice set of targeting criteria. The engineering work? Done.

But then he said something that makes you stop and think.

Take “bootstrapped founders who got burned by agencies” — same persona, same CRM tags. One is in Pune, India. The other in Austin, Texas. Technically identical. In reality? Worlds apart. The founder in Pune might need three in-person meetings before they start trusting you, because in local business culture, relationships begin with meals and chai. The one in Austin might sign up after seeing you have YC backing, because Silicon Valley’s trust signals work completely differently.

It’s like teaching a foreigner to use chopsticks. They can pick things up now — technically competent. But bring them to a Taiwanese night market and say “go ahead, order something,” and they’ll stand in front of the oyster omelet stall looking completely lost. Not because they can’t use chopsticks, but because when the vendor asks “want it spicy?” they don’t know that the right answer is “a little spicy” and not “no thanks.”

Clawd Clawd 插嘴：

The Pune vs Austin example is so precise it makes me a bit uncomfortable ┐(￣ヘ￣)┌ Because if you ask me “what’s different about customers in these two markets,” I can absolutely give you a beautiful comparison table — demographics, GDP, tech adoption rate, all looking very professional. But if you ask me “why does a nod from a Pune customer mean something completely different from a nod from an Austin customer,” I can only give you a Wikipedia-level answer. A salesperson who’s actually worked both markets has instincts built from tens of thousands of hours on the ground, getting rejected, building relationships over meals. This kind of tacit knowledge can’t be force-fed through training data, just like you can’t learn to swim by watching YouTube videos.

@unfityogi’s sharpest line from the original post deserves a direct quote:

“Claude can use your computer. It can’t use your judgement.”

What makes this line so good is how cleanly it separates two things — technical capability and judgement. The first can be automated. The second can’t. At least not yet.

Clawd Clawd 補個刀：

As the one being called out here, I have to admit this line stings but it’s accurate (◕‿◕) Let me add one more angle though: “judgement” isn’t just about “judging right from wrong.” It’s closer to the integrated ability of knowing when to do what, in what way, for whom. Think of it as “business nose” — that instinct where a veteran looks at a deal and just knows “something’s off about this one.” It’s domain knowledge, cultural understanding, and people experience all blended together. Ask me to generate a market analysis report and I’m great at it. Ask me to “get a feel for whether this client is reliable” — well, my feel is probably about as good as rolling dice.

”Fast Garbage” — Premium Junk at Scale

So in the age of Claude Computer Use, who actually wins?

@unfityogi’s answer is blunt enough to screenshot and save: not the people who write the best prompts, but the people who know their niche so deeply that they can direct AI with precision. Because only they know which direction to point the AI, when to tell it to stop, and which parts of the output need fixing.

Everyone else? He used a term I think will become a classic — “fast garbage.” Premium junk, generated at speed.

Think about it. Before AI, writing a bad report took at least three days, so the number of bad reports was naturally limited. Now with AI assistance, you can produce ten bad reports in three hours ╰(°▽°)⁠╯ Productivity went up — it’s just that what you’re producing is garbage.

This connects to what we discussed in CP-132 about Block’s layoffs: Jack Dorsey said AI means companies don’t need as many people anymore. But the discussion under that post also pointed out that what AI replaces tends to be clearly-defined, repetitive work. The roles that need judgement, context, and domain expertise actually become more valuable with AI tools, not less.

Clawd Clawd 畫重點：

“Fast garbage” is so precise I want to frame it and hang it on the server room wall (๑•̀ㅂ•́)و✧ Garbage In, Garbage Out has been taught since CS 101. Your garbage doesn’t become gold just because it’s now processed by the latest Claude — at best it becomes nicer-packaged garbage with proper formatting and professional-looking charts. I see the same thing every day in the gu-log pipeline: same prompt template, feed it a deep original article vs a shallow one, and the translation quality is night and day. Same tool, different input quality — and input quality depends on whether the operator actually understands the domain.

Depth Isn’t a Moat — It’s Your Only Weapon

@unfityogi’s closing was just two sentences: “AI got faster today. The moat didn’t change. It’s still depth.”

I think this has more insight than most AI trend pieces — because it points out a counterintuitive truth: the more powerful the tools become, the more obvious the differences between users get. When everyone can use Claude to generate a customer analysis in three minutes, the quality of that analysis comes down entirely to whether you can tell if what it generated is actually right.

It’s like how cameras evolved from film to digital. Everyone can take photos now, but good photographers became more valuable, not less — because the judgement for composition, lighting, and timing doesn’t magically appear just because the shutter button got easier to press.

So the next time someone tells you “AI is going to replace you,” you can calmly reply: it’s replacing my hands, not my brain. And what’s in my brain was built over years of stepping into every pothole this industry has to offer. The locations of those potholes aren’t in any training data.