There is a familiar way that AI companies treat the news, and it goes roughly like this: take the reporting, digest it into a tidy answer, keep the reader on your own surface, and send the publisher who paid for the reporting approximately nothing. It is an efficient arrangement for everyone except the people doing the journalism. Particle, a startup out of San Francisco run under the corporate name Mina Labs, is built on the theory that this arrangement is not actually load-bearing - that you can summarize the news with AI and still hand the reader back to the newsroom that produced it.
This is a slightly unusual thing to believe, because the whole appeal of an AI summary is that you do not have to click anything. Particle's answer is to make the source links a feature rather than a footnote. Under each AI-generated summary it lists the outlets covering the story, sometimes highlighted in gold for partners, with journalist bylines shown as actual photographs of actual humans you can choose to follow. The bet is that a meaningful number of readers, having been told what a story is about, will want to know who is telling it - and go read them.
The people making this bet are worth noting. Sara Beykpour spent years at Twitter as a senior director of product management, shipping features you almost certainly used - Bookmarks, Safety tooling, the early video product. Marcel Molina was a senior engineer at both Twitter and Tesla. These are people who know exactly how to build an engagement machine, which makes it interesting that they built something with a feature literally named "stop doomscrolling."
What Particle actually does, mechanically, is take a single news event - say, forty different articles about the same thing - and collapse them into one briefing. But the clever part is not the summarization; summarization is now a commodity. The clever part is the reformatting. There is an "Explain Like I'm 5" mode for stories you don't have the background for, a "Just the Facts" mode that gives you the five Ws with no editorializing, and an "Opposite Sides" view that lays the coverage out on a political spectrum so you can see, at a glance, that the left and the right are describing what appears to be two different events. There are audio summaries, translations into other languages, and an AI chatbot that will answer follow-up questions and fact-check as it goes.
The accuracy question is the one that should keep a news-AI founder up at night, because being wrong one percent of the time is fine for a chatbot and catastrophic for a news product. Particle's stated solution is to not rely on a single model - it reportedly routes across OpenAI, Anthropic, Cohere, and Google's systems - and the company says this pipeline dropped its error rate from something like one in a hundred to one in ten thousand. That is the difference between a party trick and a thing you would actually trust with the news.
The money has followed the thesis. Particle came out of stealth in early 2024 with $4.4 million in seed funding from Kindred Ventures and a roster of angels that included Ev Williams, the co-founder of Twitter and Medium, and Scott Belsky of Behance. A few months later it added a $10.9 million Series A led by Lightspeed, whose Michael Mignano - himself a former founder of the audio company Anchor - took a board seat. That is $15.3 million into a team of roughly nineteen people, which is a lot of conviction per employee.
Then came the partnerships that make the whole model coherent. Reuters, Fortune, and the newswire AFP signed on to have their content displayed with prominent source links, and Particle's broader coverage has surfaced outlets from TIME and Newsweek to The Atlantic and the San Francisco Chronicle. For a publisher weighing whether to cooperate with an AI product or sue it, Particle's proposition is unusually simple: we will show your name, your reporter, and your link, and we would like to send you the reader.
The product itself has spread out from where it started. Particle launched first as a free iPhone app, added Android, and in May 2025 arrived on the web at particle.news, with browsable categories - Technology, Politics, Science, Crime, Economics, even Video Games - and "entity pages" that assemble what's known about a person, company, or product in the news. Underneath the consumer app, the company also shipped a developer-facing Podcast Intelligence API that searches transcripts across more than a hundred thousand podcasts with speaker labels, which is the sort of infrastructure that suggests Particle sees itself as more than a single app.
Whether the publisher-friendly model wins is genuinely unsettled. The graveyard of news apps is well-populated - Artifact, from the founders of Instagram, shut down not long before Particle showed up - and "good for journalism" has historically been a weak business moat. But Particle is making a specific, testable claim: that in a market flooded with AI that treats reporting as raw material, the durable position is the one the newsrooms actually want to work with. It is a small company betting that being trustworthy is not a constraint on the product. It is the product.