This Modern Web

Introducing Feedster

I love the web. It’s the reason that I have a career, it’s why I got into design, and it has enabled so much of my (and our society’s) modern life that I cannot imagine going back to life before it existed.

A lot has been said in the last few years about the rise of AI agents and how people access content. We’re seeing a wild shift in how people get their information: search engines (and AI agents at large) are increasingly turning into answer engines, reducing real human traffic and interaction in the open web.1 In parallel, AI-generated summaries appear in search results, making users less likely to visit individual websites further pushing behavior toward answer-seeking instead of browsing.2 As AI systems consume and remix the web’s content at scale, infrastructure providers and publishers are actively reshaping how crawling and access work, reflecting how contested the old traffic model has become.3 While this behavior does allow for incredible efficiency in going deep on a particular subject, it also removes a certain serendipity that the “old web” provided, something that has long been treated as a real (and designable) part of discovery.4 Coupling this recent trend with the way social media websites have created walled gardens of content5 over the last decade, we’re left with fewer places to wander and discover.

That’s why I decided to build Feedster. I miss blogs, personal websites, and the internet publications that I grew up consistently reading online.

At its core, Feedster is a feed reader that allows its users to organize their feeds into different channels. Channels are like folders that contain many feeds. Each channel has its own top-level feed mixing all of the feeds within it. Feedster is different from traditional RSS readers in that it’s web-based and all of the feed syncing happens on the server. There are some technical workings that allow the posts to stay cached locally for faster load times, but all of the feed updating happens server-side first and Feedster operates as a “thin client”. Additionally any feed that gets added to Feedster by one user, immediately becomes discoverable by all other users of the system. I’m still working on making search really great, but even in its basic form it is nice to be able to share feeds (especially independently published) so easily.

A channel on Feedster
A channel on Feedster
Creating a public channel allows others to subscribe to it, like a shared playlist of feeds
Creating a public channel allows others to subscribe to it, like a shared playlist of feeds

The reading experience in Feedster allows users to view feed items’ content directly in the app itself. There are two views for each piece of content: a reader view, which is similar to the reader view in Safari, and a web view, which will render most websites directly inside of Feedster, so that users don’t have to leave the app to get an idea of what the real content looks like and can easily paginate through posts within a given channel.

Reading content on Feedster
Reading content on Feedster

Despite being a countermeasure to the erosion of some of the original web’s appeal, Feedster was also built to connect with AI agents. Using Feedster’s hosted MCP server, users can connect to their Feedster account and give AI agents access to the content that they follow, so long as MCP is supported on the AI agent’s client. This gives direct access to the content that users care about in near real time, unlocking many workflows for agents that may resolve on recent information without requiring the agent to search the web directly.

Having a chat with Claude about some content in my Feedster
Having a chat with Claude about some content in my Feedster
Here’s a transcript of that chat…
❯ /feedster

⏺ Feedster - front_page.get (MCP)(limit: 10)
  ⎿  ### Front Page (10)
     1. **Stolen Gemini API key racks up $82,000 in 48 hours** (id: 7391348) — Hacker News (Mar 3, 20
     26 12:41 PM)
     … +27 lines (ctrl+o to expand)

⏺ Here's your updated front page:

  1. Stolen Gemini API key racks up $82,000 in 48 hours — Hacker News 🆕
  2. College students add puppy training to their course load — Washington Post 🆕
  3. Seattle Xcoders Talks This Week Considered Unmissable — inessential.com
  4. Anthropic's popularity surges amid Pentagon rift — Semafor
  5. Introducing the AI Gateway — Vercel
  6. Pinterest shares pop 8% as Elliott Management takes $1B stake — CNBC 🆕
  7. Anthropic and Alignment — Stratechery
  8. Amazon Cloud Unit's Data Centers In UAE, Bahrain Damaged In Drone Strikes — Slashdot 🆕
  9. 'Pray We Make It Through the Night': Iran's Capital Under Siege — NYT 🆕
  10. Apple Introduces M4-Powered iPad Air — MacStories

  A few new stories since last check. Want to read any of these?

❯ Tell me about the anthropic thing and also provide additional context.

⏺ Feedster - items.get (MCP)(item_id: 7391249)
  ⎿  ### Anthropic's popularity surges amid Pentagon rift
     - ID: 7391249
     - Feed: Semafor
     … +20 lines (ctrl+o to expand)

⏺ Feedster - items.get (MCP)(item_id: 7386043)
  ⎿  ### Anthropic and Alignment
     - ID: 7386043
     - Feed: Stratechery
     … +53 lines (ctrl+o to expand)

⏺ Feedster - items.get (MCP)(item_id: 7391234)
  ⎿  ### US Defense Secretary Pete Hegseth's and Trump's actions against Anthropic have serious legal
      issues, and its designation exceeds what the statute authorizes (Lawfare)
     - ID: 7391234
     … +21 lines (ctrl+o to expand)

⏺ Feedster - items.get (MCP)(item_id: 7386043, content_offset: 4000, max_chars: 4000)
  ⎿  ### Anthropic and Alignment
     - ID: 7386043
     - Feed: Stratechery
     … +53 lines (ctrl+o to expand)

⏺ The Anthropic-Pentagon Standoff

  Here's the full picture:

  What Happened

  The Pentagon demanded that Anthropic allow its AI models to be used for "all lawful use cases" without
  restriction. Anthropic refused, drawing hard lines on two specific uses:

  1. Mass domestic surveillance — Anthropic argues AI-driven surveillance can assemble scattered data into
   comprehensive profiles of Americans' lives at massive scale, and that the law hasn't caught up with
  these capabilities.
  2. Fully autonomous weapons — Anthropic says frontier AI is not yet reliable enough to take humans out
  of the loop on selecting and engaging targets, and offered to do joint R&D with the DoD on improving
  reliability — an offer the Pentagon rejected.

  The Escalation

  - Feb 24: Defense Secretary Pete Hegseth gave CEO Dario Amodei a deadline — comply by 5:01 PM Friday,
  Feb 27, or face consequences.
  - Feb 27: Anthropic publicly refused to budge.
  - Feb 28: Trump directed federal agencies to stop using Anthropic's products. Hegseth designated
  Anthropic a "supply chain risk" — a label previously reserved for US adversaries, never before applied
  to an American company. This ended Anthropic's ~$200M defense contract and bars other defense
  contractors from working with them.
  - The Pentagon also threatened to invoke the Defense Production Act to force compliance — which
  Anthropic noted is contradictory (you can't call a company both a security risk and essential to
  national security).

  The Fallout — Surprisingly Positive for Anthropic

  - Claude shot to #1 on Apple's US App Store charts as public support surged
  - ChatGPT saw a 295% surge in uninstalls after OpenAI signed its own unrestricted DoD deal
  - Defense contracts were only about ~5% of Anthropic's total business
  - Prediction markets are betting Anthropic will be largely unaffected

  The Debate

  - Lawfare argues the designation is "political theater" — it has serious legal issues and exceeds what
  the statute authorizes. They don't expect it to survive legal challenge.
  - Ben Thompson (Stratechery) takes a more cautious view, arguing the designation "would decimate
  Anthropic" and that Anthropic's position — having an unelected CEO decide what the government can do
  with AI — is "intolerable and misaligned with reality," even if their specific concerns are legitimate.
  He draws an analogy to international law: ultimately, power decides what's allowed, not principles.
  - Scott Alexander noted the financial exposure is small and expects limited impact.

  Sources:
  - Semafor: Anthropic's popularity surges amid Pentagon rift
  - Stratechery: Anthropic and Alignment
  - Lawfare: Pentagon's Anthropic Designation Won't Survive Legal System
  - NPR: Anthropic rejects Pentagon demands
  - Axios: Hegseth gives Anthropic deadline
  - Fortune: Hegseth issues ultimatum to Anthropic
  - Washington Post: Pentagon-Anthropic fight reshapes Silicon Valley relations
  - TechPolicy.Press: Timeline of the dispute

✻ Cooked for 38s

❯ How did you know to pull that Stratechery article too?

⏺ It was on your front page from the earlier fetch — item #7, "Anthropic and Alignment" by Stratechery.
  When you asked about "the anthropic thing," there were actually three related articles across the two
  front page pulls:

  1. Anthropic's popularity surges amid Pentagon rift — Semafor
  2. Anthropic and Alignment — Stratechery
  3. US Defense Secretary's actions against Anthropic have serious legal issues — Techmeme/Lawfare

  I grabbed all three since they were clearly about the same story from different angles.

Feedster is still in its early infancy. I think there is a lot more I can design and build to bring back the serendipity of the old web. To start, I am focusing on these things:

  • Prioritize a slower type of content that isn’t constrained to 240 characters
  • Give users the ability to create sharable channels of feeds that other users can subscribe to
  • Create a better experience for curious minds that want to discover new sources of content outside of the mainstream (or social media platforms)
  • Bring RSS into 2026 and expose it to a new audience

I hope that you’ll give Feedster a spin. I’ve been having a blast building it.

Footnotes

  1. Reuters Institute for the Study of Journalism, “Journalism, media, and technology trends and predictions 2026” (discusses search evolving into AI-driven answer experiences and concerns about referral traffic). ↩

  2. Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results” (reports lower click-through rates when AI summaries are shown). ↩

  3. Cloudflare, “Cloudflare Just Changed How AI Crawlers Scrape the Internet at Large” (describes tooling/policies to block AI crawlers and the scale of the crawler dynamic). ↩

  4. M. A. Rahman, M. A. I. Hossain, and M. A. Hossain, “Exploring Opportunities to Facilitate Serendipity in Search (SIGIR 2015)” (studies serendipity in web search and frames it as a meaningful part of discovery). ↩

  5. Coding Horror, Jeff Atwood, “[https://blog.codinghorror.com/avoiding-walled-gardens-on-the-internet/]” ↩