Hacker News Daily Digest: The Best Threads, Debates, and Big Ideas in Tech

The average developer spends only about two to five percent of their time actually writing code. That one comment on Hacker News this week is the perfect, ice-cold rebuttal to the entire AI hype cycle we’ve been tracking. Last episode, we were all caught up in the launch of Mythos AI and its promise to write entire applications, but this week, the community is reminding us that the job was never about the typing in the first place. So what IS the job? If it’s not writing code… what are these people doing for the other ninety-five percent of their day? Well, according to the threads, they’re doing the one thing the models can’t. They’re thinking. One user put it perfectly: “Most of the time is spent figuring out what the right thing to do is, not writing the implementation.” And that’s the whole game right there.

It’s understanding the business problem, talking to users, designing the system, debugging a legacy mess that nobody has touched in a decade, and then—only then—translating that solution into syntax. This isn’t to say AI is useless. Far from it. The consensus is that large language models have fundamentally changed the texture of the work. One comment that really stood out said that LLMs “evaporated 90% of the ‘moments of despair’ when you have an error and googling it isn’t helping.” I think every developer knows that feeling. That pit in your stomach when you’re stuck on something stupid, some obscure configuration flag or a typo you just can’t see. AI is incredible at fixing that.

It’s a super-powered linter, an infinitely patient rubber duck. Another user pointed out that experimentation is now basically free. You don’t have to spend a week building a proof-of-concept; you can just ask the model to sketch out five different approaches in ten minutes. You converge on a solution by trying things, not by trying to predict a complex system in your head. But here’s the turn. For every moment of despair the AI solves, it creates a new, more insidious one. People are talking about the sheer brittleness of AI-generated code. It works, until it doesn’t, and then you have no idea why. There’s a comment from a developer who spent an entire afternoon trying to get three different top-tier models—we're talking the best of the best—to generate a simple CRUD API across multiple files.

Just a basic, textbook-simple application. And they all failed. Repeatedly. They’d hallucinate function calls, get tangled in their own logic, or just spit out nonsense. The developer’s conclusion? The models are great for single-file scripts, for isolated problems. The second you need context, state, or any kind of architectural thought… they fall apart. And you get this warning, this almost cynical sign-off from another user: “Just hope that's someone else’s problem when your solution becomes too brittle to use, cannot be maintained, or fails catastrophically.” Ouch. You know, this whole dynamic… it feels incredibly familiar. Where have we seen this before? A revolutionary technology comes along that promises to automate a skilled profession, causing widespread panic.

But instead of replacing the professionals, it just automates the most tedious parts of their job, forcing them to level up and become more strategic. It’s the story of accountants and the spreadsheet. When VisiCalc and then Lotus 1-2-3 and Excel came out, the fear was that accountants would be obsolete. Why would you need a human to manually calculate figures in a ledger when a machine could do it instantly and without error? But that’s not what happened. The number of accounting jobs in the U.S. actually GREW after the spreadsheet became mainstream. It didn’t eliminate accountants; it eliminated the job of being a human calculator. It freed them up from the tedious arithmetic to focus on the real work: financial analysis, strategy, forecasting, and finding the story inside the numbers.

The spreadsheet became a tool for thinking, not just a tool for calculating. That’s exactly what we’re seeing with AI and developers. The models are automating the boilerplate, the syntax lookup, the basic scaffolding. They are taking over the 2-5% of the job that is pure transcription. And in doing so, they are forcing developers to be better at the other 95%—the architecture, the product thinking, the user empathy, the systems design. The analogy isn't perfect, of course. Writing code involves a level of creative problem-solving that balancing a ledger doesn't. But the structural pattern—automating the low-level tasks to elevate the high-level ones—is identical. And you can see this shift happening in real-time if you look at what people are actually building.

The “Ask HN: What are you working on?” thread for May is a fantastic snapshot of this. It's not a thousand people building another to-do list app with AI. You see these incredibly specific, human-centric projects. Someone is building a search engine for European night trains, but the goal isn’t just to find the cheapest ticket from A to B. The goal is to suggest trip ideas. It’s optimized for discovery and experience, not just logistics. That’s a human problem, not a code problem. You see a group in London that calls themselves an LLM hacking collective. They’re not just using GPT models off the shelf; they’re actively dissecting them, fine-tuning them on niche datasets, trying to understand how they work from the inside out to use them for creative tasks.

They’re treating the AI not as a magic oracle, but as a raw material to be shaped and understood. This is the 95% work in action. It’s meta-level thinking about the tools themselves. And it’s happening alongside projects for personalized Hacker News digests—hello—and countless other bespoke tools designed to solve a very specific, very human-scale problem. And here’s the part that really brings it all home. This is the climax. The Hacker News community itself seems to be… well, getting over it. There’s an old thread from the beginning of the year, “Ask HN: What are your predictions for 2026?”, that’s still getting comments. And the predictions are shockingly sober. One of the top-rated forecasts is this: “AI stays the top story but in a boring way as novelty wears off and models get cheaper and faster… No AGI moment.” Think about that.

The people closest to the fire are predicting that the fire is about to become… boring. That AI is transitioning from a speculative, world-changing technology into just another piece of infrastructure. Like the cloud. Like databases. Essential, powerful, but ultimately, a solved problem you build on top of. The same user predicts a market shift. The money will stop flowing to companies with speculative growth and start flowing to companies that can deploy AI profitably. The hype bubble, they argue, is already deflating. And this sobriety comes with a dark undercurrent. Another prediction warns of “Digital Dementia 2.0,” a state where “people’s minds will become even lazier due to prolonged daily use of LLMs… causing severe psychological problems.” So we have this dual forecast: the tech becomes boring infrastructure, while the societal impact becomes profoundly unsettling.

And this skepticism is everywhere. A new code analysis tool called Mythos—yes, the same one from last week—is getting absolutely roasted. One user cuts right to the chase: “The big hype around this model so far was primarily marketing. I see no evidence that this setup finds issues to any particular higher or more advanced degree than other tools before.” The emperor, it seems, has no clothes. So what does this all mean? It means the narrative is splitting. There’s the external story, the one you read in the mainstream press, full of marketing hype from companies promising to replace your entire engineering department with a single API call. It’s a story of revolution, disruption, and obsolescence.

But then there’s the internal story, the one you find in the trenches of Hacker News. It’s a much messier, more complicated, and frankly, more grounded story. It’s a story about a powerful but flawed tool. A tool that removes old frustrations while introducing new ones. A tool that, like the spreadsheet before it, is automating the tedious and forcing a generation of skilled professionals to redefine their own value—moving from people who write code to people who solve problems. The community isn't scared of being replaced. They're busy figuring out how to wield this new hammer, figuring out which parts are solid steel and which parts are just marketing fluff. The conversation is no longer about who can write the code.

It’s about who understands the problem.