Tuesday, 11:40. Slack is unread in three channels that all feel important. Email has a thread with seventeen replies and no decision. Two model tabs are still generating. The dashboard shows twelve tiles marked urgent; half of them were urgent last week too.
You opened the morning with one question. Something like: Do we ship this week or kill the feature? Or: Do I hire, or stretch the team another quarter? The exact wording does not matter. What matters is that by noon the question is still open, and the room feels busy.
Busy is easy to mistake for progress.
You asked for clarity. What came back was volume: another summary, another take, another "here's a framework," another ping that resets the clock. Each new input felt like work. The original field — the one thing that would actually change what you do — got thinner under the pile.
Volume as progress
Watch what happens after the first fluent answer arrives.
Relief shows up. The plan looks clean. You almost move. Then someone drops a contrary link, or you paste the same question into a second model "just to check," or a dashboard tile blinks red and you chase it. The decision clock resets. Fresh input registers as diligence.
That loop has a name worth keeping: volume-as-progress — mistaking the accumulation of takes for movement on the one thing that would change the outcome.
It feels responsible. You are reading. You are comparing. You are "staying informed." The afternoon fills with adjacent cleverness: angles that are interesting, true in some other context, and unable to answer this week's question. By evening you have a thicker folder and the same open choice.
Two familiar distortions help the loop stay alive.
Confirmation bias — the tendency to search for, interpret, favor, and recall information in a way that confirms prior beliefs — shows up when new prompts mostly hunt agreement. You ask the model to steelman the plan you already like. You open the Slack thread that will praise the direction. You skip the customer call that might kill it.
Anchoring — leaning too hard on the first piece of information when later judgments adjust around it — shows up when the first model answer becomes the room's gravity. Every later take is "agree / tweak / push back" relative to that first fluent page. The first page was never sworn in as evidence. It was just first.
Neither bias requires a villain. Both thrive when volume is the metric. Loud recent takes own the room because they are available; older, quieter facts that would settle the question stay unopened.
The cost is not "too much information" as a slogan. The cost is a shrinking field. The one question that would change the week gets buried under inputs that cannot answer it.
The signal cut
Before the next search, the next model round, the next "quick check," install a cut.
Three lines. Paper or a sticky note is enough. A blank message draft works. Fancy tools are optional and often become more volume.
- Write the one question that, if answered, would change what you do this week. Not three questions. Not a research agenda. One. If you cannot write it in a sentence, you are not ready for more input — you are ready to clarify the question.
- Name three inputs that cannot answer it. Mute or park them. The dashboard tile that is always red. The Slack channel that argues process while the product question sits. The third model tab opened for comfort. Park means park: you are allowed to return after the cut, not during it.
- Open only what can falsify or confirm that one question. A metric. A customer call. A contract clause. A verse or a number you can open. A person who was in the room when the last similar bet failed. If you cannot name what you are opening, do not open another feed.
Public language for the move: cut to the field that matters.
The cut is structural. It changes what is allowed into the hour. It does not require a new personality, a digital detox weekend, or a manifesto about attention. It requires a written question and a short list of things you will not let steal the next forty minutes.
If line one is blank, stop. Do not treat another chat transcript as the decision. Clarify the question first. Volume will not write it for you.
What quieter and clearer looks like
You will know the loop broke when the evidence is boring and checkable.
- The input list next to the same decision got shorter, on purpose.
- A one-question card survived the afternoon without being replaced by a "better" question every hour.
- A parked thread stayed parked. You did not reopen it "just in case" before the check finished.
- Someone asked "what are we actually deciding?" and the room could answer in one sentence.
You will not know from vibes. Feeling calmer is nice; it is not proof. Proof is a shorter field and a question that still stands when the feed gets loud again.
Practice: the one-question card
Before another model round this week, fill this:
The only question that changes this week: Inputs that cannot answer it (park): What I'm opening once:
Rules of use:
- If the first line is blank, do not open another tab.
- If the park list is empty, you have not cut yet — you are still collecting.
- If "what I'm opening once" is another opinion feed with no openable object, rewrite it until the object is real.
Run the card once on a real sticky choice. Time-box the open (thirty to ninety minutes). Then decide, defer with a date, or change the question — with the card still visible.
The card is the practice. The essay is only the reason to pick it up.
Same question, one standard
When you do take the one question into chat, keep the question stable across models. Do not let a second model rewrite the question into a more flattering one. Ask the same sentence. Compare what holds.
On askaxiom.ai the product intent is a governed multi-model relationship — one standard across models — with teaching on /library when you want the operating language. The durable move is the card, with or without the product.