InApp : Tech Radar

The US government just pulled the plug on foreign access to Fable 5 and Mythos 5. At 5:21pm ET today, Anthropic received an export control directive citing national security authorities. The move is blunt. It suspends access for any foreign national, including our own employees, regardless of where they're located. The weirdest part is that the government didn't actually explain why. There are no specific details in the letter about what the national security concern is. It's a sudden, opaque curtain call for a set of models we spent thousands of hours red-teaming with the UK AISI, the US government, and various private firms before launch. We did the work. We invited the regulators in. We spent weeks trying to break the safeguards so we could fix them. Now, despite that collaboration, the government has decided these models are too risky to leave in the hands of non-US citizens. It makes you wonder what exactly they found during those red-teaming sessions that shi...

The only way to get a human to actually listen to you now is to prove you spent time on the request. We're drowning in AI-generated noise. When a teammate sends over a block of code or a debug summary, the first thing most of us do is scan for the tells of a LLM. If it looks like a raw prompt output, we subconsciously value it less. It's a weird new etiquette problem. On one hand, an AI that has deep access to our internal codebase and docs can produce something genuinely useful in seconds. It's efficient. But forwarding that raw output to a colleague feels like a shortcut that signals a lack of effort. We've reached a point where the quality of the answer matters less than the perceived work that went into it. So, where do we draw the line between being productive and being lazy? The Signal-to-Noise Crisis LLMs have made the cost of generating "perfect" communication zero. When everyone can produce a polished, professional email in two seconds, th...

For years, installing a third-party tap in Homebrew was essentially a leap of faith. You were trusting that the maintainer wasn't hiding a malicious Ruby script in their repository, because once you added that tap, Homebrew would happily execute that code on your machine. It's a massive security blind spot that's always felt a bit reckless, especially for anyone managing a production environment. The latest update finally fixes this with a new tap trust mechanism. Now, Homebrew won't just blindly evaluate code from an untrusted tap. It flags them first and requires you to explicitly trust the source before anything runs. It's a sensible change, and honestly, it's overdue. There is plenty of other stuff in this release, like sandboxing on Linux and initial support for macOS 15. But the real story is how Homebrew is finally tightening the screws on how it handles external code. I'm curious if this will actually change how people manage their taps, or ...

We stopped treating the browser as an application platform and started treating it as a document viewer. The result was that our growth metrics exploded. It sounds counterintuitive, especially if you've spent the last decade building complex SPAs, but focusing on HTML-first doubled our user base literally overnight. This wasn't a quick win. We had two previous attempts to solve this problem, and both were expensive failures. In the most recent disaster, we paid contractors overseas to build a full React app. It was technically "modern," but it was a ghost town. The mistake was thinking that a heavy client-side framework was the only way to handle the complexity. We were so focused on the "app" experience that we forgot how the web actually works. I want to show you exactly where we went wrong and why stripping everything back to basic HTML actually made the product better. The JavaScript Fatigue Point The heavy-client approach failed because it s...

Anthropic is now silently nerfing Claude if it thinks you're using the model to build a competing AI. They aren't giving you a warning or a "policy violation" popup. They've just implemented interventions that degrade the model's performance specifically when you ask about pretraining pipelines, distributed training infrastructure, or ML accelerator design. It's a strange move. Most companies just block a prompt entirely if it violates a safety guideline. But this is different. This is a quiet degradation of quality, a deliberate choice to make the tool less effective without telling the user why. I've seen this pattern before in other parts of the stack. Plenty of companies are moving away from generic APIs to build their own custom embedding and reranking systems to avoid this kind of dependency. I did the same thing for my own projects, training my own reranker because relying on a third party for core logic is always a gamble. The real qu...

Anthropic just cut the price of their high-end models by more than half. Fable 5 and Mythos 5 are now $10 per million input tokens and $50 per million output tokens. It's a massive drop from the Mythos Preview pricing, and it's a move that tells us exactly where the industry is heading. We're seeing a pattern where "state-of-the-art" is becoming a commodity. Anthropic claims Fable 5 beats everything they've ever released in software engineering and scientific research. That might be true, but the real story isn't the benchmark scores. It's the fact that high-end intelligence is getting cheap enough to actually use at scale without bankrupting a project. I'm skeptical about how long these margins can last, but for now, the barrier to entry for heavy-duty LLM workloads just collapsed. The question is whether the performance jump in Fable 5 actually justifies the switch, or if we're just chasing numbers on a spreadsheet. The new pricing...