The internet is not the problem. The feed format is.

Is the internet the problem, or is it the feed?

The internet is not the problem; the feed is. Specialist and primary sources have never been more abundant or easier to reach. What makes staying informed exhausting is the format most people use to reach them: an infinite stream, ranked for engagement, that shows the same event many times. The fix is to go to sources directly and let a system read them.

That position is narrower than it sounds. It does not defend every corner of the web, and it does not say the reader is at fault. It says the bottleneck sits in one specific layer, and that layer can be replaced.

What does the internet actually offer someone who needs to stay informed?

The internet offers direct access to the places where changes are first recorded. Complaints about information overload tend to skip this part, so it is worth stating plainly: the raw material of good intelligence is richer and more open than it has ever been.

Consider what a reader can open today without a subscription or a gatekeeper:

• Filings and regulatory records. Annual reports, insider transactions, merger documents and rule proposals, published by the issuer or the regulator itself.
• Preprints. Research appears before peer review and before press coverage. arXiv alone recorded 24,226 new submissions in a single month, a record at the time.
• Company channels. Changelogs, engineering blogs, pricing pages, job listings and executive posts, which often show a strategy before an announcement names it.
• Practitioners who write. Engineers, analysts, clinicians and operators who publish newsletters, threads and long talks about their own field.

These sources share two properties. They report changes early, and they report them with detail that coverage strips out. A newsroom summary of a filing is shorter than the filing for a reason.

What sits between readers and those sources?

Between most readers and those sources sits the feed: a single scrolling surface that collects items from many places and decides their order. The social timeline, the news app, the aggregator homepage and the overloaded inbox are all versions of the same format.

The feed format has three properties that matter here:

1. It is infinite. There is always another item below the last one. The format has no concept of "done."
2. It is ranked. Order is set by recency or, more often, by a prediction of what will hold your attention.
3. It is duplicated. One event arrives as the original post, the reposts, the reactions, the articles about the reactions and the threads summarizing the articles.

None of these properties came from the sources. A filing ends. A changelog has a last entry. A specialist publishes once a week. The endlessness, the ordering and the repetition are added by the layer in between.

Why is ranking for engagement a poor fit for staying informed?

Ranking for engagement is a poor fit because it optimizes for interaction, and interaction is a different quantity from importance. A post can be widely discussed and change nothing; a dry regulatory notice can change everything and attract no replies.

The mechanics are public in at least one case. X released the code of its For You timeline, and the documentation of its ranking model explains that the model outputs a set of engagement probabilities: the chance you will like a post, repost it, reply, click into the conversation, or react negatively. Those probabilities are combined in a weighted sum. In the weights published with that code, a predicted reply counted 27 times as much as a predicted like, and a reply that the author then engaged with counted 150 times as much.

That design is coherent for its purpose. A platform wants conversation, and the model rewards posts that start it. The consequence for a reader trying to stay informed is structural:

• Posts that provoke rise above posts that inform.
• Commentary on an event outranks the event's primary record, because commentary invites replies.
• What you see depends on what people like you reacted to, not on what changed in your subject.

No amount of discipline from the reader changes the objective the ranking serves.

Why does a better feed not solve the problem?

A better feed does not solve the problem because it keeps the unit of delivery the same: individual items, which the reader still has to turn into an understanding of what happened. Most proposed improvements fix one property of the format and leave the other two in place.

• Chronological timelines remove engagement ranking. They keep the infinite scroll and every duplicate.
• Muting and curated lists reduce noise from people. They do nothing about one event arriving from fifteen accounts you do follow.
• RSS readers restore direct subscriptions to sources, which is a real step forward. They still present an unread count that grows faster than anyone can clear it.
• Algorithmic "quality" filters change what ranks first. They do not tell you when you have seen enough.

The common thread is that each version still shows you more. The reader remains the deduplication engine, the relevance filter and the writer of the mental summary. The future of information is not a better feed, because the feed's basic contract, here is another item, is the part that needs replacing.

What does going to sources directly look like?

Going to sources directly means choosing where changes are first published, reading those places for what changed, and accepting a written account as the output instead of a stream. It reverses the feed's three properties: a fixed list instead of an infinite stream, relevance instead of engagement, developments instead of duplicates.

For a reader following one company, the source list might be its newsroom, its changelog, its regulatory filings, two executives who post, and one analyst who covers it well. That is six sources, all primary or close to it, with a finite amount of new material each week.

The obstacle has always been the reading itself. Six sources for one company becomes sixty for a sector, and nobody checks sixty places every morning. That labor is why feeds won: they did the collecting, even if they collected badly.

The labor is now the part that software handles best. A system can read every new item from every chosen source, group the items that describe the same event, discard what does not bear on the reader's subject and write a short account of what moved. Side by side, the two models look like this:

• A feed delivers items, ordered by recency or predicted engagement, never ends, repeats events and leaves the source behind the summary.
• An intelligence system delivers developments, ordered by relevance to one reader, ends when the news ends, merges repeats and links each line to its source.

A feed shows you more. An intelligence system shows you less, on purpose. Kindal is built on the second model: you choose the sources, and it writes a brief from the sources you trust only when something in them changes.

What are the limits of going direct?

Going direct has real limits, and the main one is discovery. A fixed source list only knows what it was told to watch, while feeds, for all their flaws, are good at surfacing voices and topics you did not know to look for.

Three other caveats deserve weight:

• Some feed content is primary. A founder announcing a shutdown on X is the original record, not commentary. The line between source and feed runs through platforms, not around them.
• Source choice can narrow your view. A list built only from people you already agree with produces a tidy, confident and incomplete picture.
• A system is only as good as its list. Software that reads well cannot recover an event none of the chosen sources reported.

The practical answer is to separate two jobs that the feed merged. Use feeds, sparingly and on purpose, to discover new sources. Use the source list, read by a system, to know what changed. Review the list every few months and promote the voices that discovery turned up.

What changes when sources and feeds are separated?

When sources and feeds are separated, the internet stops looking like the problem. The same web that produces endless timelines also produces the filings, preprints and expert writing that make informed judgment possible. The difference is whether you reach that material through a layer designed to hold attention or through one designed to report change.

The abundance was never the enemy. The format in between was. Choose the sources, let something tireless read them, and the internet becomes what it was supposed to be: the best library anyone has had, read on your behalf.