As AI takes over the internet, Palo Alto-based search engine Kagi is bringing its handpicked collection of non-commercial, human-authored websites to mobile devices through new “Small Web” apps for iOS and Android. The “Small Web,” in Kagi’s definition, includes sites created by individuals, like personal blogs, webcomics, independent videos, and more.

These are the types of properties that formed the basis of the early web, before it became dominated by ad-supported business models and platforms controlled by large corporations. They’re also increasingly the kind of sites that can be harder to discover on today’s web, where so much content is infused with, if not directly authored by, AI.

The search startup first launched its idea for a “Small Web” initiative in 2023, designed to promote this kind of content in its search results and through a dedicated website. In March, the company announced it’s expanding these efforts with browser extensions, mobile apps, and a way to filter results by category.

The Small Web website is like a modern-day StumbleUpon as it randomly displays one of the selected sites, then lets you click a “next” button to move to another. Like StumbleUpon, the goal is to help users discover the parts of the web they might otherwise have missed.

With the addition of categories, users can now limit discovery to just those topics of interest from the more than 30,000 “Small Web” sites in Kagi’s index.

These are also available in Kagi’s new mobile apps for iOS and Android and its browser extensions. Here, you can select what sort of content you’d like to see, like videos, blogs, code repositories, or comics. You can also view a list of recently viewed or popular sites, and read them in a distraction-free mode. Plus, you can save your favorite sites and articles to return to later.

While the initiative to make less-trafficked parts of the indie web more visible is a worthy one — especially at a time when AI-generated content is masquerading as human creation — some Kagi users complain that the Small Web product isn’t going far enough.

On the discussion forum Hacker News, one person pointed out that Kagi is limiting its selection to sites with RSS feeds that have recent posts, ruling out unique, single-purpose websites or experimental pages from being included in Kagi’s collection. Another was frustrated when they came across a supposed “Small Web” site that sounded suspiciously like it may have been written with AI.

Still, the concept of a human-curated web of content that’s also written by humans could be something worth building, especially if Kagi’s original concept of becoming a Google alternative by offering a premium, paid search engine doesn’t pan out.

In the meantime, people can suggest new sites for the Small Web via its GitHub page.

Apple has published its first “background security improvement” update to patch a security bug in its Safari web browser on iPhones, iPads, and Macs.

According to a new security advisory posted Tuesday, Apple said a security researcher discovered a bug in WebKit, the browser engine that powers Safari and other apps. The bug, if exploited, could allow a malicious website to potentially access data from another website in the same browser session.

Apple explains that background security improvements are “lightweight” software updates that contain important fixes for security vulnerabilities, which the company pushes to customers’ devices in between larger software updates.

These updates, which debuted with iPhones, iPads, and Macs running the latest version of iOS, iPadOS, and macOS (ver. 26.1 and higher), can contain fixes for certain software components, such as Safari, its WebKit engine, and other system libraries that benefit from occasional ongoing security updates.

Apple did not give a reason for why it patched this specific bug, and a spokesperson for Apple did not immediately comment when contacted by TechCrunch.

When we downloaded the new background security update, it only required a quick device restart, rather than the longer reboot typically reserved for software updates containing more substantial fixes.

Prior to Tuesday’s first background security improvement, Apple published several security fixes to software testers to trial the new update feature before it launched.

Y Combinator’s famed CEO Garry Tan told a SXSW audience that he’s got “cyber psychosis” and is barely sleeping because he’s so excited to be working with AI agents.

“I sleep, like, four hours a night right now,” he told his interviewer, fellow VC Bill Gurley, during an onstage interview Saturday. “I have cyber psychosis, but I think a third of the CEOs that I know have it as well,” he joked about his current AI obsession. (At least, we hope he was joking. AI-induced psychosis can actually be a dangerous condition.)

“Once you try it, you’ll realize: It’s like I was able to re-create my startup that took $10 million in VC capital and 10 people, and I worked on that for two years, and I took anti-narcoleptics — I remember, you know, sort of being on modafinil,” he described, referencing the sleep-preventing drug that’s popular with the startup hustle-culture crowd. (Tan sold his Y Combinator-backed blogging startup Posterous to Twitter back in 2012.)

But now, his psyche is so amped working with AI agents, it’s a natural insomnia.

“I don’t need modafinil with this revolution. Like, I’m up. I slept at 4 a.m. I woke up at 8 a.m.,” he said. “I wanted to sleep more, but I couldn’t because: Let’s see what’s going on with the 10 workers. I’ve got like three different projects going right now.”

He’s so excited about his agents that on March 12, just two days before the interview, he proudly, freely shared his Claude Code (CC) setup on GitHub under an open source license. The setup included six “opinionated” Claude Code skills he developed. Skills are reusable prompts stored in special “skill.md” files that instruct the AI how to behave in specific roles or tasks.

“I’ve been having such an amazing time with Claude Code, I wanted you to be able to have my *exact* skill setup,” he posted on X. He called his Claude Code setup “gstack.”

Since then he’s added several more skills. The gstack GitHub repository currently lists 13, but it seems like every hour Tan tweets about something new.

In one post, he gave an example of how his setup works. First, he gets Claude’s opinion on whether a startup idea or feature is a good idea using a skill where Claude acts like CEO. He uses another skill to have Claude write the feature as an engineer, and another to review its own work for bugs and security issues as a code reviewer. Other skills cover design, documentation, and so on.

The love for gstack began immediately: His tweet went viral on X and trended on Product Hunt. It’s accumulated nearly 20,000 stars on GitHub with 2,200 “forks,” meaning people who have taken the files to modify for themselves.

But shortly after releasing gstack, Tan posted a tweet that caused a heap of hate, too.

He wrote that a CTO friend told him gstack was “god mode” that instantly found a security flaw in his company’s code and predicted it will be widely used.

To quote just a few of the many hater comments that followed: One founder posted to X: “(1) Garry should be embarrassed for tweeting this. (2) If it’s true, that CTO should be fired immediately.”

Vlogger Mo Bitar did a take on gstack called “AI is making CEOs delusional” in which he pointed out that the project was essentially “a bunch of prompts” in a text file. The vlogger summarized the common complaint: Developers who use Claude Code already have their own versions of this.

Added one person on Product Hunt, “Garry, let’s be clear and honest: if you weren’t the CEO of YC, this wouldn’t be on PH.”

So who’s right? Is gstack a uniquely useful way to work with Claude Code? Or unremarkable? To find out, I asked the experts, including Claude (which, not surprisingly, absolutely loved it). I also queried ChatGPT and Gemini, both of which were surprisingly positive.

Gstack is a group of “reasonably sophisticated prompt workflows, but they’re not ‘magical,’” ChatGPT opined. “The real insight here is that AI coding works best when you simulate an engineering org structure. Not when you just ask: ‘build this feature.’”

Gemini called the setup “sophisticated,” adding that “gstack is essentially a ‘Pro’ configuration. It is less about making coding easier and more about making it correct.”

Claude called gstack “a mature, opinionated system built by someone who actually uses it heavily,” adding, “It’s one of the better examples of Claude Code skill design out there.”

We’ll take that as a thumbs-up from an expert on the topic.

On Monday, Tan explained in another X post, “I took modafinil just to stay awake longer to be able to turn the momentary crystalline structures I had in my brain into lines of code before sleep or human distraction turned it to grains of sand. I love coding but I love coding with AI even more. I speak it listens and we create. I see the structure and it is built. There is no more powerful an experience to me than that.”

Tan did not respond to multiple requests for comment.