Agentic news is news consumption delegated to AI agents. You define your sources, topics, and format once; software agents then monitor those sources continuously, read everything they publish, and compose a personal briefing for you on a schedule. The agent works from your standing instructions, not from an engagement-maximizing algorithm.
That last sentence is the whole point, so it is worth unpacking. For fifteen years, the dominant way to encounter news has been the algorithmic feed: an opaque ranking system that decides what you see, optimized for how long you stay. Agentic news replaces the feed's objective with yours. The AI still does enormous amounts of automated work, reading hundreds of articles you never could, but the definition of "relevant" comes from instructions you wrote and can edit, not from a model of what keeps you scrolling.
How is agentic news different from an algorithmic feed?
Both use machine intelligence to cope with infinite content, so the comparison is fair. The difference is who sets the objective and whether you can inspect it.
| Algorithmic feed | Agentic news | |
|---|---|---|
| Who sets the goal | The platform | You |
| What gets optimized | Engagement and time-on-app | Coverage of your stated topics and sources |
| Source selection | Decided for you, mixed with ads | Chosen by you, publisher-direct |
| Can you read the rules? | No | Yes, they are your own instructions |
| Delivery | Endless scroll, always on | A finite edition, on a schedule |
| Typical failure mode | Doomscrolling and rage-bait | Blind spots you forgot to instruct for |
Neither is magically neutral. An agentic briefing has exactly the blind spots of your instructions. The difference is that your blind spots are visible and editable, while a feed's incentives are not.
How does an agentic news pipeline work?
A serious agentic news product runs a content pipeline before any writing happens. At Avos, and in the category generally, the shape is: agents fetch new items from every configured source overnight, follow through to full article text, deduplicate the same story reported across many outlets down to one, rank what survives against your stated interests, and only then write summaries grounded in the fetched text with links to the original reporting. Finally the pieces are composed into a finite edition, typically delivered by email, because an inbox delivery on a schedule is the opposite of an infinite scroll: it arrives, it ends, and it does not ping you between issues.
The grounding step is what separates agentic news from generic "AI content". The agents summarize journalism that exists and cite it; they do not generate news from a language model's memory.
What stays human in agentic news?
More than you might expect. First, trust: deciding which outlets deserve to be your inputs is human editorial judgment, and agentic tools make that decision more consequential, not less, because everything downstream is built on it. Second, the beat: you decide that your briefing covers, say, European energy policy, your industry, and your local city, a combination no algorithm would ever assemble for you. Third, correction: when a briefing over-covers something or misses a nuance, you say so and the instructions change.
A useful way to think about it: agentic news does not remove the editor, it makes you the editor-in-chief of a staff of tireless readers. The judgment stays yours; the reading does not have to.
Where does Avos fit?
Avos is agentic news in its most literal form: a personal newsletter written by AI agents. You brief an AI editor in plain language about the topics, companies, and markets your edition should cover; overnight the agents read your trusted sources, including RSS feeds and market data, and every morning you get one private edition, in your language, for a single reader: you. Editions are built from typed blocks (news headlines, article briefs, stock and crypto data, custom AI sections), each with its own sources, and delivery runs on your schedule and timezone in any of six languages.
If you want to see what a finished agentic briefing looks like before building one, the template gallery has live sample issues, from a general daily brief to markets and tech and business editions. For the RSS-specific angle, see What is an agentic RSS reader?
FAQ
Is agentic news the same as AI-generated news?
No. AI-generated news usually means articles written by a model, sometimes with no underlying reporting, which is how you get error-ridden content-farm sites. Agentic news is the opposite arrangement: AI agents read real journalism from sources you chose and summarize it with links back to the originals. The AI does the reading and condensing; the reporting stays human.
Will an agentic briefing trap me in a filter bubble?
It reflects exactly the instructions you give it, which cuts both ways. Unlike an engagement algorithm, the "bubble" is a config you can read and edit: add sources you disagree with, add a wildcard topic, tell the editor to include perspectives from outside your usual outlets. The bias is explicit and yours to change, which is the best defense any news system currently offers.
What do I need to get started with agentic news?
A list of sources or topics you actually care about, and about ten minutes. With Avos you describe your beat in a chat, pick a schedule, and the first edition arrives the next morning. Starting from a template is faster still: pick one close to your interests and edit it by messaging the AI editor.