InApp : Tech Radar

Anthropic just gave us a knob for Claude's brain. For the first time, you can explicitly tell the model how much computational effort to put into a specific task. It's a subtle shift, but it changes the relationship from "send a prompt and hope for the best" to actually managing the model's reasoning process. The release includes a lot of other updates, too. The new Claude 3.5 Sonnet shows improvements in coding and agentic workflows, and the team ran the usual alignment checks to make sure it isn't behaving erratically. Most of the benchmarks look solid, showing better performance on practical knowledge work and reasoning compared to the previous version. I'm curious to see if this extra compute actually translates to better results for complex tasks, or if it's just another way to burn through API credits. We need to find out if being able to dial up the effort actually makes the model smarter, or if it just makes it slower. The mechanics ...

The idea that being available 24/7 is a sign of dedication is a lie. It’s actually a leading indicator that your architecture and your processes are broken. If you have to be constantly firefighting, you haven't built something resilient. You've just built something that requires a human sacrifice to keep running. We're being told that AI is about to 10x the productivity of the white-collar workforce. The math is pretty simple, even if the implications are messy. If these tools actually work, I should be able to produce my entire week's worth of output by Monday at noon. This creates a massive tension in how we measure value. If the work itself becomes cheap and fast, what happens to the person who is still billing by the hour or measuring success by the number of emails sent? We need to figure out if we're actually getting more efficient, or if we're just creating a new kind of digital burnout. The myth of the heroic developer The "hero develop...

Last.fm is finally out from under the thumb of a media conglomerate. After years of being tucked away inside a larger corporate structure, the service is officially operating as a standalone entity again. It’s a move that feels less like a corporate restructuring and more like a long-overdue separation. I’ve watched Last.fm go through enough hands to know that being part of a larger ecosystem usually means your roadmap is dictated by someone else's quarterly earnings. When you're a niche piece of infrastructure, you tend to get swallowed by whatever broader strategy the parent company is chasing. Seeing them reclaim their own identity is interesting, mostly because it leaves us wondering if they actually have the resources to survive on their own. The big question is whether this independence actually changes anything for the people who still use the site every day. We'll have to see if they can actually build a sustainable business without the safety net of a larg...

We're hitting a wall with LLMs. The idea that we can just "chat" our way through complex workflows is starting to feel like a fantasy. As these interfaces become more common, I've noticed the friction of crafting the perfect prompt often outweighs the actual value of the output. It's a lot of cognitive overhead for a result that frequently misses the mark. I saw this clearly last week when I stumbled upon several GitHub repositories that were actively spreading malware. I tried using an AI agent to help me figure out the best way to report and mitigate the spread, but the response was useless. It gave me generic, high-level advice that didn't help me take any real action. It was the same experience I had years ago working as a developer, asking a business owner a direct question about a task and getting a response that completely bypassed the technical reality of the problem. The tech is getting better at mimicking conversation, but it's still failing ...

The Appeal of Talking to AI The appeal of talking to an AI comes from how quickly it turns a thought into a response. You speak or type a prompt and get an answer back without having to hunt through menus or learn a special syntax. That immediacy feels like having a knowledgeable partner who is always ready to help. Early adopters often mention speed and convenience as the biggest draws. Writers use voice prompts to draft emails or brainstorm story ideas while their hands stay free. Students ask the model to explain a math problem step by step, getting a tutoring session that adapts to their pace. Some people simply enjoy casual chat, finding the AI’s replies a low‑pressure way to pass time or work through thoughts. Under the hood, the system relies on a language model that generates text token by token. Latency depends on model size and the hardware it runs on; a 7‑billion‑parameter model on a modern GPU can return a reply in 200‑400 ms, while larger models may take a second or tw...

ASML’s machines don’t just make chips — they make modern life possible. The Dutch company builds the only tools on Earth capable of etching the impossibly fine patterns onto silicon that power everything from smartphones to AI servers. When a single supplier holds that kind of leverage, it’s not just a business story — it’s a geopolitical fault line. Governments are now treating ASML like strategic infrastructure, not a commercial vendor. Export controls, subsidy negotiations, and quiet diplomatic pressure have turned its headquarters in Veldhoven into an unlikely cockpit of global tech rivalry. The U.S. wants to slow China’s chip ambitions. The Netherlands wants to keep its crown jewel. And China? It’s pouring billions into building alternatives, knowing that without access to ASML’s extreme ultraviolet lithography systems, its semiconductor dreams hit a hard wall. What happens when a company that sells $200 million machines becomes the choke point in a tech cold war? And mor...