"An AI agent that keeps up with your field for you" is the kind of sentence that sounds great and means nothing until someone tells you what actually happens when it runs. And a lot of the time, what happens is disappointing: the tool runs a keyword search and emails you the results. That's not an assistant. That's the exact firehose you were trying to escape, now arriving on a schedule, with a friendlier subject line.
A genuinely useful agent has to do something harder than fetch. It has to judge — to look at a few dozen candidates and throw most of them away for defensible reasons, then explain the ones it kept. Fetching is easy; judging is the whole job. Here's the real sequence a Folio Discovery Agent runs each time it wakes up, with the specific failure each step is there to prevent.
On its schedule, the agent fans out across scholarly sources — and, in deep mode, the live web — for recent work matching what you told it to watch. It excludes anything already in your library.
Why the order is the point
Walk it backwards and you can see why each step needs the one before it. Deliver is worthless if it dumps a hundred items on you — so it needs Brief to give the run a shape. A briefing is hard to trust if you can't see the parts — so it needs Explain to say why each finding earned its place. Explanations are wasted on noise — so they only run on what survives Rank. And ranking only works if there's a real pool to rank — which is what Scan is for, fanning out across scholarly sources and, in deep mode, the live web.
Pull any step out and the chain breaks in a predictable way. Scan without rank is the firehose. Rank without explain still makes you open every abstract to check the ranking was right. Explain without brief gives you twelve good sentences and no sense of the week. The sequence isn't decoration; it's the difference between a feed you trust and one you learn to ignore.
The hard step, up close
If one step decides whether you keep using the thing, it's ranking. Anyone can pull a list of recent papers on a keyword. The hard part is ordering them so the ones that matter to you float up — and "matter to you" is doing a lot of work in that sentence.
A Folio agent scores each candidate against an interest vector: your stated topic, blended with the actual papers already in your library, so the ranking reflects what you keep, not just what you typed. Then it weighs that relevance against how cited a paper is and how recent it is. Those three pulls trade off constantly — the freshest preprint is barely cited; the most-cited classic is years old. Play with the balance and watch the order rearrange:
Toggle what the agent should weigh and watch the order change. Real agents blend all three.
There's no single right setting, which is exactly why personas exist — a Vanguard leans hard on recency, an Archivist on citations. The agent picks a sensible blend for its style; you can lean it whichever way your project needs.
Where you stay in control
None of this is a black box you have to take on faith. You set what the agent watches and how aggressively it scans. Every finding carries the one-line reason it was kept, and you're free to read it and disagree. You can ask the agent follow-up questions about what it surfaced, or tell it to look harder, or narrower, or further back.
The line we care about is simple: the agent does the watching and the first-pass judging; the decisions that carry weight — what to read closely, what to cite, what to ignore — stay with you. It's assistance, not autopilot, and the difference shows up precisely in moments like ranking, where the machine can sort the pile but you decide what the sort means.
If your field moves fast, point an agent at the narrow question you actually work on, set it to run daily, and let the morning briefing carry the weight of keeping up. You'll spend your attention on the papers that matter instead of the search for them — which was always the point.
Folio's Discovery Agents scan, rank, explain, and brief — so you decide instead of dig. Join the waitlist.