Hacker News Daily: The Hottest Tech Debates & Breakthroughs, Curated Fast

DeepSeek just released a native coding agent that uses a clever caching trick to make API calls dramatically cheaper. Last week on the Hacker News digest, we were talking about the big debates that define the community, and today, one of those debates is about the very code we write—and just how much it costs to write it. This isn't just a small optimization; it’s a fundamental change in the economics of building with AI. And that's just our lead story. Also lighting up the boards today... OpenAI's co-founder Greg Brockman did an interview where he basically mourned the death of old-school tech media. He remembers when computer magazines had actual code listings in them, and now he says it's all just lawsuits and insider corporate drama.

He's not wrong, is he? The scale of money and power has transformed what we even talk about when we talk about tech. Then there's the health debate that will not die: seed oils versus sugar. A cardiac specialist posted a warning that the current "seed oil panic" is actively hurting his patients. Why? Because it’s a distraction. People are so focused on avoiding canola oil that they're ignoring the much bigger, scientifically proven enemy in their diet: sugar. It’s a classic Hacker News thread—a mix of biochemistry, public health, and deep skepticism about modern food. The core question is, when you only have so much discipline, what’s the right thing to focus on?

And finally, in the monthly "Ask HN: What are you working on?" thread, we get a glimpse of the other side of Greg Brockman’s coin. While the giants fight over market cap, the community is still... well, tinkering. There's a new macOS app called Drawers for managing project windows. There's a guy in London who's spent six months building a network of LLM hackers just to push the boundaries of what's possible, calling it a "guided hackathon." It’s a reminder that even if the headlines are all about billions of dollars, the actual heart of the industry is still someone in a text editor trying to build something cool. Okay, let's go back to that DeepSeek agent, because the technical detail here is what makes it so important.

If you’ve ever worked with AI agents, you know they can be expensive. They make a lot of API calls, and each call costs money. The obvious solution is caching. If the agent asks the same question twice, just give it the answer you already have, right? The problem is, it almost never works. Here's the catch that the DeepSeek team pointed out in their docs. Most caching systems, especially for agents, have a hit rate below twenty percent. That's... terrible. It's barely worth implementing. And the reason is that agent loops are messy. On each turn, the agent might rewrite the context, reorder its thoughts, or even just inject a fresh timestamp. Any tiny change to the request means the cache sees it as a brand new query.

The exact byte prefix doesn't match, so... cache miss. You pay again. What DeepSeek did is find a way to stabilize this process. They figured out how to preserve the prefix of the request so that the cache actually works. It sounds small, but it’s not. It’s the difference between a tool being a neat toy and being a viable, scalable production system. Competitors like OpenCode are reportedly struggling with this. Their users complain about cache stability and tool calls that break the whole system, driving up costs. This is a pattern we've seen again and again in tech. It’s not about the big, flashy feature. It’s about the quiet, boring, infrastructural improvement that unlocks everything else.

Think about it. This is the same shape as the move from owning your own servers to using AWS. The headline was "rent computers from Amazon," but the real revolution was in elasticity and utilization rates—boring stuff that changed the economics of starting a company. Or go back further. Think about the invention of just-in-time compilation for programming languages. Suddenly, code that was "too slow" became fast enough, and whole new types of applications became possible, all because of a clever trick happening under the hood. The analogy isn't perfect, of course. A compiler optimization is universal. This DeepSeek cache trick is specific to their platform.

But the shape of the innovation is the same: a low-level, almost invisible change that has massive downstream effects on cost and capability. It makes building with AI cheaper, which means more people can do it, which means we'll see more experimentation. It’s a quiet change that could have very loud results. Now, let's connect this to that other major thread—Greg Brockman's lament and the seed oil debate. It seems like a total context switch, but it’s not. Both of these stories are about the same thing: the economy of attention. Brockman is looking at the tech industry and saying, our attention is in the wrong place. He says, "I remember when computer magazines were aimed at programmers and had code listings in them.

Then it became all about lawsuits and insider corporate drama." He’s pointing out that the narrative has shifted from making things to fighting over the things already made. The scale of money, he says, is "beyond imagination" compared to the old IBM versus Microsoft days. And that money acts like a massive gravitational force, pulling all of our collective attention toward quarterly earnings and legal filings instead of clever caching techniques. Now hold that thought, and look at the seed oil debate. A commenter, a cardiac specialist, says, "The seed oil panic is hurting my cardiac patients. Humans only have so much attention and discipline." There it is.

That's the link. He’s saying his patients come in, worried to death about soybean oil or canola oil—these modern, refined oils that are, yes, ultra-processed. But in their panic to eliminate them, they’re not replacing them with olive oil and fresh vegetables. They're just doubling down on other processed foods that don't have "seed oils" on the label but are loaded with sugar, which is a far more established villain in heart disease. As one commenter put it, "Replacing 'seed oils' with hamburgers and french fries fried in tallow won't magically help your health." The structural pattern here is identical to Brockman's point. In a world of infinite complexity, you have a finite budget of attention.

Where you point it determines the outcome. Brockman is worried the tech world is spending its attention budget on corporate soap operas instead of engineering. The cardiologist is worried his patients are spending their health budget on a trendy "panic" instead of the boring, proven advice of "eat less sugar." Where have we seen this before? Everywhere. It's the core problem of the modern information environment. We are constantly being pushed to focus on the novel threat over the familiar one, the dramatic narrative over the statistical reality. The fight isn't just about what's true; it's about what gets to be the headline in our own minds. Is the real enemy the thing that's new and scary, like refined seed oils or a rival's new AI model?

Or is it the boring, chronic problem that’s been sitting there all along, like sugar consumption or technical debt? So while the engineers at DeepSeek are finding breakthroughs by focusing intently on a single, technical problem, the rest of the world—in tech and in health—is getting pulled into a battle over what we should even be paying attention to in the first place. So this week sets up a fascinating tension. We see the power of intense, focused, technical work creating real value, like with DeepSeek's agent. And at the exact same time, we see these powerful voices, from inside and outside the industry, warning us that our collective focus is drifting.

We're being distracted by drama, by panics, by the sheer scale of the money involved. The question this week leaves us with isn't just about what's new. It's about what matters. The most valuable resource isn't capital or code; it's focused attention. And the biggest fight is for control of it.