TECH BLOG

Let me explain why I started looking for a token usage monitor in the first place. When I’m working on something, I usually go all in on the topic. A fragile mental web tends to form in my head from bits of various information: the task itself, chunks of code, overlapping ideas from a pile of open tabs, documentation, saved articles, answers from different LLMs, and so on. And then, right in the middle of all this, usually at the worst possible moment, your main LLM tells you your quota is gone.

For eight years, MacUpdater quietly did one of the most thankless jobs in macOS: tracking outdated apps and letting you update them with a single click. Roughly 6,800 apps were tracked in real time, with version data stored for around 100,000 more. Then, on January 1, 2026, CoreCode wound the project down.

If you still have MacUpdater 3.5 installed, it will keep working. CoreCode released a final free version with all Pro features unlocked, and they plan to keep the database server running through the end of 2026. But the daily maintenance work that made the tool actually useful has ended. False positives are already starting to show up, and it will only get worse from here. So for everyone who likes keeping their apps current, the question of what replaces MacUpdater is very much on the table.

AI started running businesses this month, not just assisting them. March was about AI starting to act. April is about AI starting to manage. Agents are now negotiating contracts, running on custom silicon, and being baked into your terminal before you even open a file. The “AI as a chatbot” era is over.

Seven things worth your attention.

1. Harness Engineering is the real job now

Everyone predicted the death of the software engineer. Then Anthropic accidentally leaked 512,000 lines of code for their Claude Code CLI. What was inside is what makes this interesting: not a thin wrapper around a model, but a massive scaffolding system with memory management, self-healing query loops, and multi-agent coordination layers (including something called the KAIROS background daemon), all designed to stop the model from collapsing under its own weight.

February was loud with announcements. March was quieter, but what actually shipped matters more than what got demoed on stage. The theme of the month: AI models are no longer just answering questions. They are clicking buttons, writing code, deleting emails (more on that later), and running on hardware that would have seemed absurd a year ago.

Seven things worth paying attention to.

1. AI now uses your computer

For years, the bottleneck was integration. If your software didn’t have an API, AI couldn’t touch it. That’s changing.

Cloudflare recently released Dynamic Workers into open beta - fast sandboxes for AI agents. The company claims they run 100x faster than traditional containers when executing AI-generated code. That’s a serious claim. Let’s see what’s behind it.

Containers Are Too Heavy for AI Agents

Modern AI agents don’t just answer questions, they can write and execute code on the fly. An agent might generate a small script to call an API, process some data, or complete a task. That code needs to run somewhere. For security reasons, it’s best to run it in a sandbox, isolated from the rest of the system.

I’ve been using Obsidian for quite a while now. Started and dropped it about three times. But this time it’s here to stay, and it’s gradually becoming part of both my work and personal workflows. There’s still a long way to go before I use it truly efficiently. What draws me in is that everything is stored locally in plain Markdown files, it’s rock solid, syncs across devices easily and quickly (using Syncthing), and has a massive plugin ecosystem. If you haven’t tried it yet, I’d recommend starting with the official website.

If you’ve ever tried to display tabular data on a Hugo site, you know how quickly things go sideways. Say you have a perfectly good CSV file with app rankings, price comparisons, or benchmark results. To turn it into a nice table on the page, you’ll either have to wrestle with Markdown markup or write HTML by hand. Both options are mediocre, and both scale poorly.

I ran into this often enough that I eventually built a proper tool, a shortcode called csv-table. You point it at a CSV file and get a styled, sortable, responsive HTML table. One line in your content, zero manual markup.

There’s this amazing app called Raycast that successfully helped me replace several paid utilities. I’ve been using it for a long time, and it’s become so integrated into my daily routine work on MacBook that it’s now just muscle memory. It’s strange that Apple couldn’t bring their Spotlight up to this level, but that’s another topic.

If you’re curious why specifically Raycast, then read the article Alfred vs Raycast vs Monarch. There are useful tips on using Raycast there.

Face recognition is a key technology powering countless solutions: from smartphone unlocking to complex security systems. Today’s market offers dozens of models, each promising high accuracy and speed. But how do you navigate this diversity and choose the right one for your project?

Whether it’s a mobile app with real-time face recognition or a surveillance system with high accuracy requirements, choosing the right model is critical. Studying comparative tests and reviews of leading face detection and tracking algorithms is the first step toward effective technology implementation. This approach will not only save resources but also achieve stable and fast performance in real-world conditions.

The explosive growth of large language models has created a new challenge for developers: how to efficiently manage access to multiple LLM providers without getting tangled up in different APIs, pricing models, and authentication methods.

Two popular services have emerged to solve this problem - OpenRouter and LiteLLM. Both tools promise to simplify LLM integration, but they use fundamentally different approaches. This article will help you understand which solution works best for your needs.