Based on this morning’s edition of The Installer from David Pierce at the Verge mentioning Notion, Craft, and Obsidian, I think the next big AI productivity tool is an AI productivity tool to help you choose a productivity tool! 😂
Based on this morning’s edition of The Installer from David Pierce at the Verge mentioning Notion, Craft, and Obsidian, I think the next big AI productivity tool is an AI productivity tool to help you choose a productivity tool! 😂
It’s cool that Microsoft is using a variety of models (I really like Anthropic’s models these days) but I’m a little surprised this deal doesn’t include hosting their models on Azure, too.
New AI-powered summarize mobile app from Kagi available on iOS and Android.
One thing I love about this is it works with YouTube videos - just share the YouTube page to the app and it pulls the transcript and gives you a nice summary so you can decide if you want to watch.
I’ve been looking for an opportunity to learn ActivePieces, an automation tool similar to IFTTT, Power Automate, or Zapier. Recently, I found the perfect use case to explore its capabilities.
This fall, my son is starting at a new school. Like many educational institutions, they use multiple communication channels to keep parents informed. They maintain a news page that’s updated frequently, but it lacks proper organization; there are no individual articles or rolling history. The school simply updates this single page with the latest information.
While I could check the page daily for updates, I thought a little automation project would be more fun!
I began by installing ActivePieces along with a local instance of ChangeDetection.io, both running via Docker on a Raspberry Pi. The setup process was straightforward, with plenty of documentation available online.
I then configured ChangeDetection to monitor the school’s news page. For those unfamiliar with this tool, ChangeDetection is an application that periodically checks web pages for updates. When changes are detected, it can send notifications or trigger various actions. It’s commonly used for price monitoring and similar applications.
In my setup, ChangeDetection monitors the school webpage on a regular basis. When it detects changes, it triggers a webhook I created in ActivePieces, sending the UUID of the updated ChangeDetection record.
On the ActivePieces side, I’ve configured a webhook endpoint that receives this UUID. The workflow then makes two API calls back to ChangeDetection: one to retrieve the current version of the page and another to fetch the previous version. This comparison approach becomes important in the next step.
Both versions are stored as variables within the ActivePieces flow for processing.
Here is where I used generative AI to make the school updates a bit easier to read. Using the “Ask AI” piece in ActivePieces, I construct a prompt that includes both versions of the web page and instructions to write a summary of the differences.
I send this prompt to Anthropic’s Claude API, which returns a concise summary of only the changes. This means I receive just the new information without having to scan through the entire webpage.
For the final step, I integrated this system with my existing reading tools. I use Feedbin to do most of my online reading. Feedbin is a service that aggregates RSS feeds and email newsletters into a single spot. It provides a dedicated email address for newsletter subscriptions so they don’t clog my inbox.
ActivePieces composes an email containing Claude’s summary along with a direct link to the original page, then sends it to my Feedbin email address. As a result, whenever the school page updates, a new item appears in my Feedbin account with an AI-generated summary of the changes. If I need more detail, I can click through to view the full page.
This solution might be somewhat over-engineered - convincing the school to adopt a proper blogging platform might be simpler. However, this project provided an excellent opportunity to explore ActivePieces and kick the tires a bit with a simple automation or dare I say “agent”.
Model Context Protocol (MCP) is a standard developed by Anthropic (makers of the Claude AI chatbot/model) that allows LLMs to interact with tools so that AI services can do things in addition to talking about things. It is still early days, but I ran across this cool app called HyperContext that allows Claude and other compatible services to search data from your mail, calendar, contacts, etc, and even take actions from Shortcuts.
If we take sponsorships, ads, and other junk out, what would a modern search engine look like? What features could be developed if users are the customers, and not the sponsors? A startup called Kagi is working on this with features to make searching fun and quick again such as:
the ability to focus your search on different types of websites - for example, to search forums or discussion boards, or on sites by independent creators rather than large media firms
adjust the search to favor certain sites or completely block others
automatically compiling all of those “listicle” sites that show “Top 10 whatever ” into a short, easily bypassed list
The whole thing is designed to make searching fast - find what you need and quickly move on.
Here’s the tricky part though. Building and running all that is expensive. Rather than sponsorships, Kagi makes money the old fashioned way: they charge for use of their service. There are a few plans; they start at $5/month. If Kagi wants to keep making money, they need to make their service really good for their end users - and it is hard to compete with the “free” service from Google. But this means their needs and ours as users - a good, fast, private search engine - are aligned.
Neat little app for sharing travel plans and such with “IRL” friends. I like the distinction here between social media and social networking
I have wondered why Apple hasn’t done this yet, given how long it has taken individual states to sign on to using digital IDs. Adding your passport to get through TSA and other ID checks can be done once and works for everyone.
The cool ElevenReader app now has a feature to upload a PDF or submit a URL, and it will generate a podcast episode about it. I think Google NotebookLLM was the first to do this, but nice to have the feature in a standalone app.
I recently discovered Listen Later. It is similar to a read-it-later app like Instapaper or Goodlinks, but in this case, it uses text-to-speech to read the articles to you. You email it an article or a URL and it shows up in a private podcast feed so you can listen to the article later. I spend a lot of time in the car and this is a great way to catch up on reading!
Back in the days of the PalmPilot, there were a few makers of little keyboards that folded up for easy use while traveling. This cool keyboard from ProtoArc is a modern equivalent and even has a trackpad. It supports 3 different Bluetooth devices and works great with my iPad mini and the Apple Vision Pro!
Lot of cool AI apps out these days, but I’m enjoying Cleft Notes. The idea is you ramble on about whatever, it records you, transcribes the text, and then organizes it into well-written notes.
I’ve never really used any custom launchers on the Mac, but recently tried out Raycast and am hooked. I left cmd-space to use Spotlight, so option-space opens Raycast. It is super quick to open an app just by starting to type its name, but the really fun part is when you start to add extensions…they have everything from adding reminders to querying ChatGPT all from one text box. Pretty handy.