AI Screening Agent or Stock Screener? What Each One Can Actually See
The short answer. A stock screener applies filters to a table of numbers and returns whatever survives. An AI screening agent reads past the numbers: it opens the filing, checks whether a metric means what it appears to mean, and can exclude or rescue a company for a reason no filter could encode. The screener is faster and completely predictable. The agent is slower, more expensive, and the only one of the two that can catch a company whose numbers look wrong for a good reason.
Both exist on this site. That is not hedging: they solve different problems, and knowing which one you are using is most of using either well.
The mechanical difference, in one table
| Screener | Screening agent | |
|---|---|---|
| What it operates on | A table of pre-computed metrics | The table, plus the filings behind it |
| How a name is excluded | It failed a threshold | It failed a threshold, or it failed a judgement |
| Speed across 3,000 names | Under a second | Minutes to hours |
| Cost across 3,000 names | Effectively zero | Real money, every run |
| Reproducible | Exactly, forever | Approximately, and that is a genuine cost |
| Catches a one-off distortion | No | Sometimes |
The row people underrate is the last but one. A screener run twice gives the same answer twice. An agent may not, and anybody telling you that is a solved problem is selling something. It matters most where you would least like it to: two people screening the same universe on the same morning can get lists that differ at the edges.
The same 3,000 names, by what each approach costs to run
Seconds, on a log-ish scale of patience. The screener is effectively free and instant. The reading step is neither, which is the only reason anybody still ships a screener, and a good reason.
Source: Seconds per full pass at 64 parallel workers
What a filter cannot see
Here is the concrete case, and it is not exotic. A company sells a division. The proceeds land in one quarter. Trailing earnings jump, the price to earnings ratio collapses, and every value screen in the world now lists a company whose actual operating business did not change at all.
A screener cannot see this. A filter on the ratio has no access to the reason. An agent that opens the cash flow statement and finds a large one-off inflow can flag it, and the good ones do.
The inverse happens as often and is more expensive: a genuinely improving business takes a writedown, its trailing earnings go negative, and every quality screen in the market drops it on the day it became interesting.
The same universe, screened two ways
An illustrative pass over a 3,000 name universe on identical value criteria. The agent's list is shorter and its composition is different: it removes distorted cheapness and rescues names a trailing metric had disqualified.
The part every screener hides: what the filters cost you
Stacking filters feels like rigour. It is also subtraction, and the thing being subtracted is not evenly distributed. Each additional hard cut removes companies that failed that one metric for a reason the screen cannot see, and good companies fail single metrics all the time.
What five filters actually do to a market
Drag the number of filters. The survivors count is the number every screener shows you. The two cells beside it, good names kept and good names cut, are the ones that decide whether the list was worth building.
The lesson is not "use fewer filters". It is that strictness is a trade with a price, the price is invisible in every screener interface ever shipped, and a screen returning eleven names out of three thousand should worry you rather than impress you.
Why a name gets cut, on a typical value screen
Only the first slice is the screen doing its job. The three below it are companies removed by an artefact of how a number was computed rather than by anything true about the business.
And then there is the bill. A reading step over a whole market is a budget line rather than a feature, which is the single best predictor of what any tool in this category will actually offer you.
Why nobody reads the whole market every day
Where a full pass actually spends its time. The model calls are the bill; the fetching is the reason it is slow.
Set the universe to what you would want covered and the checks to what a trustworthy read takes. The monthly figure is the reason the universe you are offered is smaller than the one you asked for.
Where ours is today
Our screener is fast, deterministic and covers we currently hold 7,992 live companies, of which 3,065 carry a rating and 2,871 carry a discounted cash flow estimate. It is the thing to use when you know what you are looking for.
Our screening agent sits on top of it and is honestly mid. It reads the statements behind a shortlist and catches the one-off distortion case reliably. It is much weaker at the rescue case, because rescuing a name requires understanding why a charge was taken, and that judgement is exactly where a shallow plan runs out. It also will not currently tell you it is unsure; it will give you a softer sentence instead, which is a worse failure and one we are fixing rather than defending.
The version in build changes two things: the agent chooses its next source based on what the last one said rather than walking a fixed list, and every claim it makes carries a confidence you can see and sort by.
What this cannot tell you
Neither tool tells you whether a company is a good investment. A screen is a way of deciding what to read next, and the reading is where every actual decision gets made. Anyone selling a screen as a decision has skipped the only part that was ever hard.
The bottom line
Use the screener when you know the shape of what you want; use a reading step when the numbers look too good, because too good is usually an accounting artefact rather than an opportunity. If you only take one thing: run the funnel above with five filters, look at the good names cut, and then go and look at what your own saved screens are throwing away. The screener is free and it does not need an account to try.
Educational information, not financial advice. Where this page states a figure about our own coverage it is read from the database as the page loads, so it is today's number rather than the day this was written.
One well-researched article at a time. No spam, unsubscribe in one click.
No spam, no selling your address, unsubscribe in one click. The tools stay free either way.
Keep exploring: browse the stocks we cover or see what the super investors hold.