AI writes, AI scans, and the human chair stays empty. [Image via Gemini]

We're in a funny spot with AI at the moment. As people debate its value, with one side arguing it's the greatest invention since fire, all while the other calls anything it touches pure slop, AI is actively finding a foothold in evolving how business is done, in all facets. To claim otherwise is ignorant at this point.

Software is being written more quickly with AI. Security gaps are being both exploited and solved more quickly. All matter of communication is being authored and adjusted more quickly. As I said in my first post after multiple years' hiatus, we will need to find out the best place for AI and humans to coexist, and I truly believe the discussion of AI will eventually fade to the background, just like our flareups about social, mobile, local, and whatever transformative buzzword at the moment happened to be. This is not to undercut the actual value and impact of AI, but to optimistically suggest we as humans are great at adjusting.

Our eyes and our minds have been trained since the day we were born to understand what humanity looks like. Since we all first started to learn to read or watch TV, we have absorbed what it means to be "real", and we visibly react when something feels synthetic. We all have a visceral reaction when special effects seem too special, an ad seems too perfect, or a visual defies comprehension. Something feels "off". But as the models train and get better, learning through feedback and continuous reinforcement, their images and video get better, and so does their writing.

Techmeme is aflutter this weekend about a Bloomberg report that shows where we are in this particular cat and mouse game. As AI writing becomes more human, more humans can get outed for sounding too much like AI. The end result is predictable, with people intentionally making mistakes, and writing in a way that bots won't, so they don't get accused of shortcutting their work, or leaving the hard work to an LLM. If you think the bots can't adjust to make typos on their own, or to follow instructions to sound even more human so even the bots will fail the AI detection test, well, you haven't thought this through. Maybe you've even seen the viral posts that say Mary Shelley's novel Frankenstein would be considered full AI generated because the models themselves were trained on the book.

Since kicking off my first posts here two decades ago, I've had more than 3,000 posts and countless words hand typed into a myriad of open CMS windows. I've written buckets of emails or social posts and Slack messages all over the place. But I also know that in 2026, it might make sense to let the bot do a first draft, especially when it solves the blank page problem. What I do worry about, eventually, is if the AI is doing the writing, and the AI is doing the distilling and the summarizing, what if no actual humans consumed and fully took in the content? We are all very busy people, and if not, we all prefer to be seen as such, and we fill the edges of our time with things we'd rather be doing than reading text slop.

Running a script against your content to hide em dashes or add typos to seem fallible isn't going to automatically flip the lever that confirms you're a human and the AI is not. It takes an entire body of work, a corpus that no machine has, with a real personality that's adjusted through experience. 

Nearly twenty years ago, I wrote that your blog is your brand, and later, that blogging remains the foundation for who you are online. The latest generation has grown up with blogs being too slow, perhaps, with TikTok and Instagram being more visual quick hits of dopamine and long form being less valuable. Now Gen Z is approaching college and a daunting workforce where AI has always been expected and accepted, even as its place is evolving and being debated. Some of them are definitely writing their course papers with AI. Some aren't. But the reckoning won't come from somebody getting outed for asking ChatGPT, Claude or Gemini to be their scribe. It will come when everyone is distracted by something more exciting and nobody was reading anyway.

Disclosures per always: I work at Google and use our Gemini and Antigravity tools a lot. I wrote this post by hand. But how would you know?

When I debuted Webnerd.ai last weekend, the goal was pretty straightforward: bring back the fast, high-signal, human-curated morning reading ritual I loved in the heyday of classic RSS, updated for how the open web actually works in 2026. But just as much as I've enjoyed returning to that routine,  I've appreciated steering the roadmap with Antigravity. A decade plus ago, I'd point to companies that blogged weekly with their updates and bug fixes as leaders in their space, communicating to readers what was new. Now, I can make multiple updates daily, or make fixes in minutes myself -- and have even debuted a release notes section you can subscribe to from right within the app. It's an RSS inception, basically.

Over the last few weeks of living inside Webnerd.ai whenever possible, curating my shared stream and pushing links out to LG Stream on Bluesky, every "I wish it did this" or "I wonder if it could..." moment has turned into working code rapidly. Sean Goedecke, whose blog has rapidly become one of my favorites of 2026, published a piece today titled "You can just ask AI to do things", and honestly, there is no better summary of the world we live in now. If you can think it, you can build it.

Because things have moved fast since the weekend, here is a quick highlight reel of new features in Webnerd.ai right now.

First-Class Open Identity: AT Protocol OAuth, PDS Sync, and Standard.site

If you're going to build a home for the open web, you shouldn't lock people's reading identity behind a single provider. Current employer aside, I started with Google sign-in because that's what I knew. But in fast response to public request,  Sign in with Google now sits side-by-side with native Sign in with AT Protocol. Whether you use a bsky.social address or verify with your own domain like louisgray.com, you can jump straight in without extra modals or app-password gymnastics.

Even better, your decentralized identity isn't just a login badge, it carries your reading graph. Inside the OPML & Subscriptions hub, you can still drag and drop a classic XML file, or you can hit Sync PDS to pull decentralized subscription lists straight from your Personal Data Server across tools like Skyreader, Standard Reader, and Standard.site. Think about that for a second. Am I an expert on those? No. Could I work with Antigravity to build support here? Yes. So I did.

Webnerd.ai also now speaks native at:// and standard:domain publications directly in the Subscribe dialog, and when you share a story from a Standard.site publication, Webnerd automatically writes a cryptographic site.standard.graph.recommend endorsement record to your PDS (and cleans it up if you unshare). Open standards work best when apps actually interoperate instead of just talking about it or seeing an acronymn as a checkbox.

Rescuing Newsletters from Inbox Purgatory

As much as I appreciate newsletter platforms like Substack and Beehiiv quietly keeping RSS alive under the hood, plenty of great writers still send email-only newsletters, and in my view, an email inbox sandwiched between flight confirmations, calendar invites, and shipping receipts is not the best place to read long-form writing.

You can now open + Subscribe and switch to the new Personal Newsletters tab, and you'll find your own dedicated @webnerd.ai address. Use it when signing up for newsletters (or forward existing ones via Gmail filters or the included one-click Cloudflare Email Routing snippet), and Webnerd automatically spins up a separate, clean feed for each sender inside your Newsletters folder. Newsletters belong alongside the rest of your reading, not competing with work email.

Listening to Your Stream & Real Power-Reader Triage

Sometimes even my eyes need a break from screens, or I want to catch up on my feeds while walking the dog. Hitting the new Listen button in the top bar (or on any individual post) launches a docked audio player right at the bottom of your stream.

If a feed item includes a native podcast or audio enclosure (flagged with an Audio badge), Webnerd plays the original audio directly; if it's a written post, it seamlessly reads the article aloud via a spoken Text-to-Speech queue with 1x to 2x playback speeds, an Up Next drawer, and automatic mark-as-read as each story finishes.

And because anyone subscribing to 100+ feeds may eventually face inbox-style bankruptcy, I've gone deep on power-reader ergonomics:

• Granular Mark-as-Read: A new split dropdown lets you clear items Older than 1 day, 3 days, 1 week, or 2 weeks, or Mark above as read as you scroll — so you can declare peace on Tuesday's news while keeping this morning's stories untouched.
• Folder & Subscription Management: You can now add, rename, drag-and-drop, delete, and merge folders inline or from a dedicated Manage Subscriptions view, toggle Newest / Oldest sort order, and pull Full Text readability views on stub feeds that only publish a sentence or two in XML.
• More Classic Themes: Alongside Cloud Nine, Warm Sepia, and OLED Black, desktop nostalgia nerds can also switch into Mac Classic, Mac OS X Aqua, BeOS, and Green Android.

AI on Your Terms — Never Between You and Your Feeds

I wrote earlier this month about how filtertainment eviscerated our informed monoculture. The whole point of an RSS reader is that you choose the sources, and nobody hides a post  because an algorithm decided outrage would keep you scrolling longer.

Where AI does shine in a reader is synthesis on demand — when you explicitly ask for it.

Tap Briefing in the toolbar (or hit B on your keyboard) and Gemini 3.8 Flash synthesizes an executive summary across your unread stream — scoped to your Last 100 items, Last 200 items, Last 12 hours, or Last 24 hours — clustering breaking storylines and citing the exact outlets in your subscriptions. Every expanded card also has an optional one-tap TL;DR button when a 4,000-word essay drops five minutes before your next meeting. AI acts as your research assistant, never your gatekeeper.

Mobile Overhaul, Instant Startup, and Eating Our Own RSS Dogfood

Under the hood, Webnerd.ai now boots through a single-round-trip bootstrap endpoint paired with a local snapshot cache for sub-16ms first-frame paint — so opening the tab feels instantaneous even on spotty mobile Wi-Fi.

On phones, we replaced cramped desktop menus with a thumb-friendly 4-tab mobile layout (Feeds, Search, Shared, and Folders), edge-swipe gestures, and a guided Add to Home Screen full-screen web app setup on iOS while I continue working on a native Android companion app - coming soon.

Finally, you can't build an RSS reader and not publish your own changelog via RSS. Webnerd.ai now has an official Release Notes & Product Blog at webnerd.ai/releases, complete with a one-click + Follow in Webnerd.ai button and its own RSS feed at webnerd.ai/releases/rss.xml.

If you tried Webnerd.ai over the weekend, or haven't yet, take the new audio player, newsletter inbox, or AT Protocol sync for a spin. If something bugs you or you have an idea for what an RSS reader in 2026 ought to do next, ping me on Bluesky — there's a pretty good chance Antigravity and I will have it shipped by the next weekend.