I wanted a frictionless way to talk to my Mac and paste the transcript straight into ChatGPT—without paying per minute, sending audio to a cloud API, or building a full desktop app.
The interesting part isn’t that “speech-to-text is possible.” It’s that with modern coding agents, copying the shape of a popular tool is now a ~1-hour project if you keep the scope tight.
Mole is a CLI application to deep clean and optimize macOS. It aims to combine features you might know from tools like CleanMyMac, AppCleaner, DaisyDisk, and iStat into a single binary.
It includes:
deep cleaning (caches, logs, and browser leftovers)
smart uninstaller (removes apps plus launch agents, preferences, and other remnants)
disk analysis (visual disk explorer and large-file management)
live system monitoring (CPU/GPU/RAM/disk/network stats)
Install via Homebrew (recommended):
brew install mole
Run the interactive menu:
mo
Santiago revisits last year’s predictions about what the industry was supposed to look like in 2025. It’s a great example of how the saying holds up: we overestimate what we can achieve in 1 year and underestimate what’s possible in 10 years.
The gap between AI hype and reality is massive. We were given tools, yet we often treat them like magic.
At the same time, from that whole list, the point that feels the most noticeable to me is: “Writing code manually was supposed to be obsolete.” Whether online or at my job, most people have switched to writing with the help of AI models/agents, and fewer and fewer write everything manually from scratch. Still, a human has to remain in the loop - someone needs to understand, verify, test, and take responsibility for the outcome.
And that’s probably where we are right now: AI genuinely speeds up the work, but we’re not yet at a point where we can safely hand over full control of large, production systems to autonomous agents. That likely won’t change within a year, but it’s definitely an area worth watching closely.
In this X thread Aaron Levie explains how productivity growth in various fields led to an increase in the number of employees in those sectors over a few decades. He also assumes the same will happen with AI and the entire IT industry. The democratization of knowledge and skills will enable projects to get started that once seemed unprofitable
Hey there! Ever feel like you're drowning in a sea of articles, blog posts, and news updates? Yeah, me too. Wouldn't it be awesome to have a little helper that could just... poof... give you the gist?
Well, guess what? You can totally build that helper yourself! Today, we're diving into a fun project: creating your very own AI-powered Telegram bot using the magic of n8n to summarize articles for you. Ready to automate your reading list? Let's get started!
Improving the performance of a Retrieval-Augmented Generation (RAG) pipeline might seem like a daunting task, especially when there are multiple components to consider. However, if you focus on one critical piece—the embedding model—you can make significant enhancements without overhauling your entire system. In this article, we’ll walk you through our journey of fine-tuning our RAG pipeline, sharing insights and practical solutions that can help you achieve faster, more relevant search results for your copilot application.
When it comes to enhancing our RAG pipeline, the retrieval part quickly emerged as the natural point for improvement. Instead of a complete rebuild that would risk introducing new complexity and higher costs, we set out to optimize our existing embedding model. This approach allows you to boost performance by tweaking a single component rather than the whole architecture.
Recently I had a chance to test it. Yet, this simple task gave me a little surprise. Here I’d like to share how you can do it, and what can surprise you too.
In our example, we have one bucket where photos of our pets are stored. All the files are encrypted by one shared encryption key. Even though it’s a fine solution for us, we want to improve it a little bit. We need to introduce a unique key for each pet type. We love pets, and we know that in the future we want to have them more and more. Due to this, our solution should also be extendable. We don’t want to be concerned about changing the implementation once we get a new type of pet.