You cannot opt out of the algorithm. The world publishes far more every hour than any person can read raw, so something will always decide what reaches you: an engagement model, an editor, a chronological cutoff, or your own instructions. "Escaping the algorithm" therefore does not mean living without one. It means firing the one that works for a platform and replacing it with one that works for you: an explicit, readable set of rules about whose reporting you see, on which topics, in what form, and when.
There is no such thing as no algorithm
Every way of consuming news is a ranking system wearing different clothes. The feed ranks by predicted engagement. A newspaper front page ranks by an editor's judgment. Even "just chronological" is an algorithm, one that hands your attention to whoever posts most often. So the honest question was never "algorithm or not", it is "who sets the objective, and can you read the rules?" For the platform feed the answers are: not you, and no. The rules are proprietary, they optimize your time-on-app, and the one lever you have, the follow button, is routinely overridden by "suggested for you".
Once you see it that way, the escape route is obvious. You do not need less ranking. You need ranking whose objective function you wrote.
What "making your own algorithm" actually means
Your own algorithm is a short set of standing instructions, in plain language, that determines your news. Concretely, it answers four questions. Sources: whose reporting is allowed in? This is the most powerful ranking decision anyone makes, and platforms never let you make it cleanly. Beat: which topics get covered, and which are explicitly out of scope? Form: headlines, deep briefs, data; how long, in what tone, in which language? Schedule: when does it arrive, and, just as important, when does nothing arrive?
Notice what these four questions have in common: every answer is inspectable. There is no mystery model of your behavior involved, no inferred interests you never stated. If the output is wrong, you do not wonder what the machine thinks of you; you edit a sentence.
Writing yours in practice
Ten years ago, executing personal instructions like these required a human staff, which is why only presidents had one; we wrote about that lineage in our PDB story. AI agents made the execution layer cheap. With Avos, your algorithm is literally a conversation: you tell an AI editor something like the following, and that becomes the standing order your briefing runs on.
- "Cover AI infrastructure, EU energy policy, and my watchlist: ASML, TSMC, Siemens Energy."
- "Sources: FT, Handelsblatt, these six substacks, and the arXiv digest. Nothing from aggregators."
- "One skeptical deep-dive per issue on a story everyone is hyping."
- "Keep it under ten minutes, analytical tone, in German, at 6:30 every weekday."
Overnight, agents read everything those sources published, deduplicate and rank it against your instructions, and print one private edition with every claim linked to its source. Changing your algorithm is one more message. Seeing one before writing yours takes a click: the template gallery is a set of pre-written algorithms, from a daily brief to markets, that you adopt and then edit.
The difference you feel
An algorithm you wrote optimizes for being finished, and that changes the texture of being informed. The briefing arrives, it covers your beat from sources you trust, it ends, and the system goes silent until the next issue. No pull-to-refresh, no "suggested for you", no ambient guilt. The feed's question is "what will keep this person here?" Your question is "what does this person need before their day starts?" You will notice how different the answers are within a week. The companion pieces to this one are Disconnect from doomscrolling, but stay informed and Agentic news for agentic people.
FAQ
Isn't hand-picking my sources just codifying my own bias?
Yes, and that is still the better deal, because your bias is now written down where you can confront it. A feed's bias is invisible, denies being one, and optimizes for provocation. A source list is auditable: you can see the gap ("nothing from outside my politics", "no primary sources") and fix it with one instruction, including deliberately subscribing to outlets you disagree with.
Do I need to be technical to build my own algorithm?
No. The instructions are plain sentences, not code. If you can describe what you want to follow to a colleague, you can brief an AI editor. The technical part, fetching, deduplicating, ranking, and summarizing across dozens of sources nightly, is exactly what the agents exist to do.
What about serendipity? Feeds surface things I did not know I wanted.
Feeds surface what engages people like you, which is a narrow and profitable kind of serendipity. Real discovery survives the switch fine: keep a wildcard section in your briefing ("one great story outside my beats"), keep a weekend long-read block, and keep the humans, newsletters and friends, who were always your best recommendation engine anyway.