0:00 Andrew Kelley, the creator of the Zig programming language, just declared that Anthropic is, quote, “actively campaigning to end software engineering.” Last week’s episode touched on the relentless pace of AI breakthroughs, but this week the conversation on Hacker News turned sharply to who, exactly, those breakthroughs are supposed to be for—and who they might be coming FOR.0:24 Kelley’s post wasn't just a stray comment; it was a lightning rod, and it perfectly frames the skepticism that defined the entire week’s debate.0:33 So let's get into the other headlines that are feeding this mood.0:37 The biggest community thread of the day, with almost seven hundred points, was a simple request: an "Ask HN" for a flag to identify AI-generated articles.0:47 The anxiety is palpable.0:48 People are tired of sorting the real from the synthetic, and the top comment nails the problem: "Nobody wants to label their stuff as AI generated because they removed credibility." It's a catch-22.1:01 The tool is powerful, but admitting you used it devalues the work.1:05 This isn't just a technical problem of detection; it's a social one of trust.1:10 Then you have the classic "Ask HN: What Are You Working On?" thread for July.1:15 And amidst all the weekend projects and side hustles, you get things like Uruky—a privacy-first search engine based in the EU that just added Monero for payments.1:26 It has, like, 250 monthly users.1:28 It’s tiny.1:28 But it’s a perfect snapshot of the Hacker News ethos: someone, somewhere, is always building a more private, more independent, more human-scale alternative to the giants.1:39 It's the hopeful counter-narrative to the overwhelming scale of Big AI.1:44 And just for a bit of culture, a huge thread popped up around a curated list of classic cyberpunk comics and manga.1:51 We're talking The Long Tomorrow, Akira, Blade Runner.1:54 It's more than just nostalgia.1:56 The themes are all there—dystopian megacorps, the line blurring between human and machine, resistance against technocracy.2:04 It feels like people are reaching for the old playbooks to understand the present.2:09 They’re looking for a map to a future that suddenly feels very, very close.2:14 But all of these threads—the anxiety over AI text, the hope in small-scale builders, the return to cyberpunk warnings—they all orbit the week's central drama.2:24 The one that started with Andrew Kelley's accusation.2:28 And it’s not just about Anthropic.2:30 It’s about the promises being made, the money being spent, and the first hard data that suggests the emperor might be, if not naked, then at least wearing some VERY expensive and inefficient clothes.2:43 Okay, so let's dive deep into this.2:45 Kelley’s claim is that Anthropic needs you to believe software engineering is ending.2:50 Why?2:51 Because they’ve taken thirteen point two billion dollars in investment and are apparently barreling towards a one-trillion-dollar IPO valuation.3:00 You don't get that kind of money by promising to build a slightly better coding assistant.3:06 You get it by promising to eliminate the coders entirely.3:09 You sell a revolution.3:11 This is a familiar story, right?3:13 Where have we seen this before?3:15 It’s the dot-com bubble playbook.3:17 A massive valuation completely disconnected from profitability, all floating on a world-changing narrative.3:24 You remember Webvan?3:25 They were going to revolutionize grocery shopping.3:28 They raised hundreds of millions, had a huge IPO, built massive, expensive warehouses...3:34 and then went bankrupt because the fundamental economics of delivering a single carton of milk to someone's door just didn't work at scale.3:43 They sold the future, but couldn't survive the present.3:46 That’s the shape of Kelley’s accusation.3:49 Anthropic is selling a future where their AI writes all the code, but in the present...3:54 well, the present looks a little different.3:57 And right on cue, two other threads on Hacker News provided the receipts.4:02 The first was a technical analysis from a site called systima.ai.4:06 They compared Anthropic's Claude Code to OpenAI's OpenCode on a specific task.4:11 And the numbers are just staggering.4:13 Before it even starts to read the user's prompt, Claude Code burns through about thirty-three thousand tokens of context.4:21 Thirty-three THOUSAND.4:22 For comparison, OpenAI's model used seven thousand.4:26 It's like turning on your car and it idles by burning half a tank of gas before you even put it in drive.4:32 The study's author has this killer line: "Every token of harness payload is a token of working context you cannot spend on your task." It means the model is so bloated with its own internal setup that there's less room, and less budget, for your actual work.4:49 It’s just wildly, profoundly inefficient.4:51 Now, the counterpoint is that on some tasks, Claude can batch things more cleverly, but the overall picture is one of immense waste.5:00 And if that study is the lab result, the second thread was the real-world field test.5:05 A company called Ploy.ai, which builds AI agents that create marketing websites, just published a post-mortem on their migration.5:14 They switched their production system from Anthropic's Claude Opus 4.8 to OpenAI's new GPT-5.6 Sol.5:20 The result?5:21 Their agents now run two point two times faster, and cost twenty-seven percent LESS.5:26 Think about that.5:27 For a business, that’s not a small tweak.5:30 That is a fundamental competitive advantage.5:33 They even said their own internal tests were initially biased towards Claude, because their whole system had been built around its quirks.5:41 As they put it, "Your harness is tuned to your incumbent model, and you don’t know it." Even with that thumb on the scale, the new OpenAI model blew it away on performance and cost.5:53 So what does it all add up to?5:55 You have a prominent developer accusing a multi-billion-dollar company of selling a fantasy.6:01 You have technical analysis showing their flagship product is shockingly inefficient.6:06 And you have a real customer switching away and immediately getting better, faster, and cheaper results from a competitor.6:14 This is where the dot-com analogy gets really sharp.6:17 It’s not that the technology is fake.6:20 The AI is real.6:21 It does amazing things.6:22 But the business model and the narrative built on top of that tech might be the part that's unsustainable.6:29 Webvan’s trucks were real.6:31 Their warehouses were real.6:32 The groceries were real.6:34 The thing that was a fantasy was the idea that you could do it all profitably at the scale they promised.6:41 And here’s where the analogy breaks, and where it gets more serious.6:45 Webvan failing just meant you had to go to the supermarket yourself.6:49 If a company with a trillion-dollar valuation built on the promise of replacing human labor fails, or even just pivots violently, the shockwave is going to be a lot bigger.7:00 The narrative they are selling has consequences, whether it's true or not.7:05 It changes how people plan their careers, how companies invest, how we think about the future of work.7:12 Andrew Kelley isn't just mad about code.7:14 He's mad about a story that he sees as fundamentally deceptive, being pushed with a firehose of venture capital.7:21 And this week, for the first time, the data started to agree with him.7:26 This week wasn't just another cycle of AI hype.7:29 It felt different.7:30 It felt like the beginning of a correction.7:33 Not a market correction, necessarily, but a narrative one.7:36 The conversation on Hacker News shifted from "Wow, look what it can do" to "Wait, how does it work, and what does it actually cost?" The focus moved from the magic trick to the man behind the curtain—and the discovery that he's spending an awful lot of money to pull it off.7:54 We saw it in the raw anger from a respected programmer, in the hard data of token inefficiency, and in the pragmatic business decision of a startup voting with its wallet.8:05 We even saw it in the desire for a simple label to tell us what's real anymore.8:10 This sets up a new phase in the debate.8:12 The era of pure, unadulterated hype might be ending.8:16 Now comes the audit.8:17 The questions are getting harder, the users are getting smarter, and the claims made by companies with city-sized valuations are about to be tested not just in demos, but in production, on the balance sheet.8:30 This week, the story wasn't the future AI promises to create.8:34 It was the bill for the present, and the growing suspicion that it's much higher than we were told.
