This Week's Sponsor:

LookAway

It helps you take regular breaks from your Mac to rest your eyes, without breaking your flow.

Posts tagged with "AI"

ChatGPT Is Surprisingly Thorough at Planning Running Routes→

Every now and then, I go looking for new running routes to mix things up, which was why I was intrigued by this post by Simon Willison, who gave ChatGPT Work his address and asked it to find 5K and 10K routes, using OpenStreetMap data:

It worked for 27 minutes and produced exactly what I’d asked for, as both an embedded visualization and downloadable GPX file and GeoJSON files. Here’s that 5K route:

Source: [simonwillison.com](https://simonwillison.net/2026/Sep/12/astra-running-routes/)

Source: simonwillison.com

I decided to give the same thing a try and was impressed by the results. ChatGPT spent a similar amount of time, poring over OpenStreetMap data, but it didn’t stop there. The model consulted my town’s website to account for road construction and the availability of sidewalks. Then, it visited the county’s website to review maps of greenways that aren’t part of the local road system. The final results were within 100m of exact 5K and 10K routes with notes on where to be careful to verify safe pathways and road crossings.

What I love about this sort of experiment is that it’s not something I’d ever have thought to ask ChatGPT myself. In hindsight, it’s obvious that all the data is there, ready to assemble. However, it just goes to show the many ways that I, and I’m sure many others, are still just scratching the surface of what LLMs make possible.

Permalink

watchOS 27: The MacStories Review

I haven’t been this excited about an Apple Watch in a long time. Last week, Apple introduced the Apple Watch Series 12 and Apple Watch Ultra 4, which will be available at the end of the week. The new Watches feature the S11 chip, which enables a bunch of new features, some of which won’t debut until later this year. At the same time, Apple is dropping support for more Watches than usual.

What we’re witnessing is a reboot of the Apple Watch and watchOS forced by advances in AI. The Apple Watch and its OS found their niche years ago, freeing us from pulling out our iPhones for notifications, tracking our health and fitness, and enabling the occasional Siri request. It was a good and comfortable spot for what by all accounts has been one of the most successful wearables to date, but watchOS also felt stuck in a rut.

Every year we got a few new workout types, a new watch face, and a few other features, but the OS didn’t advance much. What’s different this year is that watchOS 27 is breaking with slower hardware and looking forward to the substantially more powerful S11 chip, which will allow the Apple Watch to expand into all-new areas.

What this transition also means is that there is not just one set of features to track in watchOS 27 this year but three:

  • those that arrive today for anyone with a compatible Apple Watch,
  • those that arrive at the end of the week that are exclusive to the new Apple Watch Series 12 and Ultra 4, and
  • those that are coming later this year.

That makes this more than just a review. It’s also a preview of things to come and a roadmap for where the Apple Watch is heading.

There’s more than usual to cover this year, so let’s dive in and check it out.

Read more

macOS 27: The MacStories Review

Every OS update has a story to tell. With macOS 27 Golden Gate, that story captures a moment when Apple is faced with multiple challenges, many of which are tied to the rise of large language models.

One of those challenges is obvious. Apple was caught off guard by chatbots, overpromised a response at WWDC in 2024, and has been playing catch-up ever since.

But this isn’t simply a story of missteps or wasted opportunities. There’s more to it than that.

Read more

Logitech Releases the Developer-Focused MX Keypad for AI Automation

Source: Logitech.

Source: Logitech.

Today, Logitech introduced the MX Keypad. If it looks familiar, that’s because it’s one-half of Logitech’s existing MX Creative Console. However, instead of bundling a 3 x 3 keypad with a dialpad, the MX Keypad is just the keypad for $99.

I reviewed the MX Creative Console in 2024, and I still use it and write about ways to automate tasks with it. It’s an excellent pair of devices, so it’s not surprising that Logitech has adapted the keypad to a new use case.

Keypad, cable, and kickstand. Source: Logitech.

Keypad, cable, and kickstand. Source: Logitech.

The MX Keypad’s hardware is identical to the keypad from the Creative Console. In addition to the keypad itself, it comes with a kickstand you can use to prop it up at an angle and connects via USB-C to your Mac. My MX Creative Console review has more detail about the hardware, along with images and a short video.

I like the MX Creative Console a lot, and according to Logitech, so did software developers looking for ways to automate their AI workflows. While Logitech didn’t expressly say why it dropped the dialpad for the MX Keypad, it makes sense to me having used the MX Creative Console for months. Dialpads are great for timeline-based creative work, but less so for development work. Yes, developers might want to use the dialpad to control screen brightness or volume, but by limiting the MX Keypad to just the keypad, Logitech was able to hit a lower price point, making the device more appealing to a bigger audience.

I’ve been impressed by the number of ways Logitech has opened its devices up to automators who like to tinker, too. Similar to devices like the Stream Deck, Logi Options+ lets you build Smart Actions based on keyboard shortcuts and macros. But there’s more than that. If you’d prefer to let someone else do the automation, Logitech runs a Smart Actions marketplace with plugins tailored to certain apps. For developers there’s also a Logi Actions API for building plugins with access to each Logitech device’s full capabilities.

Source: Logitech.

Source: Logitech.

From a developer standpoint, the combination of automation options means the MX Keypad opens up a variety of possibilities like:

  • launching agents,
  • monitoring agent activity via the device’s dynamically updated keys,
  • triggering saved prompts,
  • activating voice input, and
  • automating many other repetitive tasks.

Logitech is also teaming up with GitHub Copilot as part of the MX Keypad launch, similar to the partnership that was announced with Adobe as part of the MX Creative Console. There’s a Copilot plugin available that adds Copilot controls to the keypad’s keys. Logitech customers can also take advantage of GitHub’s offer for three months of GitHub Copilot Pro+ for free.

Logitech’s MX Keypad is primarily a repackaging of an existing device with a GitHub Copilot deal thrown in to make it a better value. But it’s good to see the company addressing the developer market more directly. Having used the MX Creative Console primarily for automation and development-type work myself, dropping the dialpad from the package to lower the price is a good tradeoff. I use the dial and its other buttons occasionally, but for automation, the keypad is where it’s at, and between Smart Actions and a full plugin SDK you can point your agent at, the MX Keypad is a great choice for automators.

The Logitech MX Keypad is available directly from Logitech and from Amazon.

Perplexity Introduces ‘Hybrid Compute’ for Computer Agent on Apple Silicon Macs

Source: Perplexity.

Source: Perplexity.

Perplexity has announced hybrid compute for Personal Computer on Apple silicon Macs. Personal Computer already combined cloud-based Computer tasks with access to local files and apps, but its AI processing was primarily cloud-based. Recently, Perplexity also introduced Portable Computer, a local-first version of Computer that can escalate work to the cloud with a user’s permission.

Hybrid compute approaches the problem from a different direction. Tasks start in the cloud but if a new local classifier detects potentially sensitive information, the user is alerted and asked whether they want to route it to a local model, although users also have the option to let the flagged data be used by a frontier cloud-based model. Gemma E4B, Qwen3.6 35B-A3B, and a Perplexity post-trained version of Qwen3.6 35B are available at launch for that local work, with more models to come in the future.

Tasks that don’t require sensitive information are handled in the cloud by frontier models that also orchestrate the cloud and local tasks. Users select the cloud orchestrator and Perplexity chooses the helper frontier models used.

Another key feature is that hybrid compute works with tasks begun from an iPhone or iPad, too. Those devices could already kick off a session, but now they can also run tasks that use sensitive information locally if your Mac is turned on and running Perplexity.

Yesterday, I got a demo of the feature that walked through three scenarios:

  • an attorney researching a legal brief, with sensitive client data on their Mac, allowing them to update it locally,
  • an analyst using web-based data to update a confidential slide deck for a presentation about a company’s financials, and
  • a small business owner kicking off pricing research from her iPhone that was combined with private pricing data for her business that never left her Mac.

I haven’t tried hybrid compute myself yet, but if it works as advertised, it’s a big step forward for Perplexity, and one that a lot of its competitors are also chasing. AI agents are only as good as the tools and data to which they have access. Tool calling has been a big focus over the last year and more, but data privacy is an equally important issue that needs to be solved before more companies, especially those in regulated industries, adopt agentic tools like Personal Computer. It’s good to see Perplexity tackling this head-on.

Inside OpenAI’s Codex with Andrew Ambrosino→

On today’s episode of AppStories, Federico and I interview Andrew Ambrosino, the project lead for Codex. It’s a great conversation that covers a lot of ground, including:

  • the long history and many iterations of Codex,
  • the merging of Codex into a new version of ChatGPT based on web technologies,
  • Computer Use and Computer History, which came along with OpenAI’s acquisition of Sky last year, and
  • what goes into designing for user trust.

Then for AppStories+ subscribers, we dug into Andrew’s Mac origin story and some of his favorite Mac apps.

AppStories is available on Apple Podcasts and all your favorite podcast players, as well as YouTube. AppStories+ is a special ad-free, extended version of AppStories that’s delivered a day early in high-bitrate audio for $5/month or $50/year by itself or as part of a Club MacStories Premier membership that adds our weekly and monthly newsletters, app discounts, and other perks for $12/month or $120/year.

Permalink

Parker Ortolani’s First Impressions of the Pebble Index AI Smart Ring→

Parker Ortolani on the Pebble Index, the recently released AI smart ring:

Ambient computing is here at home on speakers and in the workplace with apps like Wispr Flow, it’s only a matter of time before we carry it with us everywhere.

The Pebble Index is the best preview of that future. The simple gesture of pressing your thumb to your index finger is unbelievably natural. Especially for quick interactions. I’ve found myself setting more reminders, recording ideas, and keeping lists up-to-date. And I know that everything is being logged on my phone in the Pebble app. Better yet, you can connect the Pebble Index to your other AI tools using MCP. There’s also no reason for skeptics to fret, it’s not always listening for hot words or logging everything it hears.

As Ortolani explains, the device ticks a lot of boxes that other AI gadgets just don’t:

The thing that I think strikes me hardest with the Index is that it’s just so socially inoffensive. It’s a form factor people already love, it’s not distracting, it’s not always listening, and it’s affordable. It could be the first truly noncontroversial AI-first hardware.

I ordered a Pebble Index months ago and it should be coming within the next week or so. I bought it for many of the features Ortolani mentions. It’s on-demand, discreet, and a form factor that doesn’t look out of place. The hardware and software matter a lot with a device like this, but if Pebble has nailed it, I can see it becoming my go-to way to jot down brief notes. I’m excited to give it a try and report back.

Permalink

Anthropic Introduces an In-App Browser for Claude Cowork

Hot on the heels of OpenAI’s addition of a cloud-based browser to ChatGPT Work, Anthropic has added an in-app browser to Claude Cowork in its desktop app. What’s interesting about the two browser implementations is that while both ChatGPT Work and Claude Cowork are meant to be used for general knowledge work, Claude Cowork’s browser is something of a hybrid that mixes aspects of OpenAI’s Codex built-in browser with its ChatGPT Work cloud-based browser.

First of all, I do not yet have access to Claude’s in-app browser, so the following is based solely on Anthropic’s announcement and other documentation. Still, the differences are worth understanding since the tools are aimed at different workflows.

Like Codex’s in-app browser, Claude Cowork’s in-app browser is built into the desktop app itself. That ties Claude’s browser to the instance of its app running on your Mac. Turn your Mac off or close the Claude app, and the browser becomes unavailable. Leave the Claude app running, and you’ll be able to control the browser session remotely through the Claude mobile app or the web.

From there, however, the browsers diverge. As you’d expect, the Codex implementation is developer-oriented, with tools to build, inspect, and annotate web pages along with access to a terminal, repos, and diffs. In contrast, Claude Cowork’s in-app browser opens automatically as needed for general-purpose web tasks instead of development tasks.

Also, while both ChatGPT Work’s and Claude Cowork’s browser implementations benefit from being isolated from your other day-to-day browser use outside those apps, their handling of login credentials is a little different. As I wrote in greater detail yesterday, ChatGPT separates the login process from the model’s activity and evaluates the login page for things like phishing schemes. Claude Cowork’s in-app browser offers two ways to log in: manually, or by importing cookies from logged-in sessions in another browser. Currently, Claude supports cookie imports on the Mac from Chrome, Edge, and Firefox, but not Safari. In addition to the security benefits that come with a browser that runs isolated from your everyday browser, Claude also requests permission before acting on a site for the first time, blocks high-risk sites, checks actions against the your requests, and uses the same prompt injection protections as its Chrome extension.

My first-run experience with ChatGPT Work’s cloud browser was a mixed bag but showed promise. Although I haven’t tried Claude Cowork’s browser yet, I’m glad to see Anthropic moving in the same direction as OpenAI here. Isolating agent browsing by bringing the browser into Claude and ChatGPT doesn’t solve every security issue, but it’s a good starting point. I expect we’ll see both Anthropic and OpenAI iterate quickly on this theme in the coming months.

Hands-On with ChatGPT Work’s New Cloud Browser Feature

Yesterday, OpenAI introduced a new ChatGPT Work feature designed to let it navigate websites that require a login without revealing your credentials to the model.

Like a lot of people, I spend far too much time clicking around websites that require a login, looking at analytics and other data, checking whether new sponsors have been booked for our podcasts, and more. It’s a tedious but necessary part of my week that slows me down and takes me away from writing and other creative work. ChatGPT’s new Work feature is designed to handle that sort of work for you.

The feature works by signing into websites using a virtual, cloud-based computer. That separates the browsing session from any browsing session on your local computer, walling the agent’s computer use off from your open tabs, cookies, browsing history, passwords, and other data. Because the login and browsing happen in the cloud, that also means work can continue whether you close the ChatGPT app or power down your computer.

According to OpenAI, an additional review model checks the sign-in request and credential destination for signs of phishing or deception. ChatGPT Work then pauses while the user signs in. Credentials entered through its secure sign-in form go directly to the cloud browser and are not visible to the model. After the user authenticates, ChatGPT resumes its work.

OpenAI also explains that its cloud-based browser doesn’t store your username and password. Instead, it saves cookies that allow you to return to a previously authenticated session. If you don’t want to remain logged in, ChatGPT’s cloud browser settings allow you to clear browser data for individual websites or all sites.

Similar to other features in ChatGPT and Codex, users have choices when it comes to which sites ChatGPT Work can access. “Always ask” is the default, requiring the agent to check with the user before using a website, but the permission level can also be set to “Auto approve” after ChatGPT checks a site for relevancy and risk or “Always allow,” which OpenAI discourages. Individual websites can also be allowed or blocked. These settings control website access, but consequential actions require separate confirmation.

ChatGPT Work's browser control works best on a mobile device and with simple login systems.

ChatGPT Work’s browser control works best on a mobile device and with simple login systems.

I gave ChatGPT’s browser use a try and the results were mixed. I wasn’t able to get the feature to work at all using ChatGPT Work in Safari on my Mac. Some websites had security measures in place that prevented me from logging in. Other times, the cloud browser asked me to take over manually, but I was unable to do so because the UI was frozen or login panels didn’t appear.

Logging into a site with a CAPTCHA required manual intervention.

Logging into a site with a CAPTCHA required manual intervention.

I had better luck using my iPhone. On one site with a simple username and password system, a login sheet appeared. I entered my credentials, was logged in, and the agent navigated the site to answer my queries. On Apple’s affiliate link dashboard website, I had to take over the browser interaction manually to satisfy a CAPTCHA, which worked after several challenges. Once logged in, the agent pulled and analyzed the data I requested. Also, after I’d signed in to both of those sites on my iPhone, I remained logged in to ChatGPT’s cloud browser, which meant I could also use them from my Mac.

In its current form, the idea of having ChatGPT Work browse signed-in websites on your behalf is better than its implementation. If you can get logged in, having an agent collect and analyze things like analytics data is fantastic. However, getting past the initial login screen is still too frustrating. That said, I’ll be keeping a close eye on the feature, which is available on eligible plans depending on rollout and workspace settings, for future use collecting and analyzing data that would otherwise require a lot of clicking around.