One day our in-house assistant simply went quiet. Not crashed. Not erroring. Just silent. Someone would type a question and get back nothing at all. Behind the scenes everything looked healthy: the question went in, the helper thought about it, an answer came back. The answer was just empty, every single time, like asking someone a question and watching them open their mouth and then politely decide against it.
The obvious move, when a helper stops talking, is to give it more. More background, more instruction, more examples of what a good answer looks like. That instinct is exactly wrong here, and it is wrong in a way most people never catch, because it does not look like a malfunction. Nothing is on fire. The helper is doing precisely what it was told: read everything, then decide whether a reply actually adds anything. And faced with an enormous pile of yesterday's conversation, it concluded, quite reasonably, that the discussion was already finished. Silence was the sensible answer. We had drowned it in context until staying quiet became the polite thing to do.
What we figured out, after staring far longer than we would like to admit, is that context is not free and it is not neutral. Every extra scrap of old conversation you hand a helper is a quiet vote for this thread is basically done. A helper trained to be useful is also trained not to babble. Pile on the history and you tip it toward keeping its mouth shut. The bug was not in the helper. The bug was in our generosity.
So we did the counterintuitive thing. We cut. We stopped shoving the entire past at it and handed it only the recent, relevant bit, plus a clear nudge that yes, a reply is wanted here. We trimmed away a clumsy, heavy go-between that had been sitting in the middle slowing everything down, and let the helper speak directly. We tuned how much room it had to work in, up a little, down a little, until it fit the real shape of the job. The headline change, the one that brought the voice back, was the smallest one and the most embarrassing to admit: less in, not more.
The lesson travels beautifully for anyone running a customer-facing operation. If you are stitching together a contacts system, a help desk, an email tool, and a chat channel, the temptation is to funnel all of it onto one screen so your team has the full picture. And then your team stops answering tickets. Not because they are lazy. Because a thread with dozens of old messages, a stack of internal notes, and a tag that says resolved-then-reopened-then-resolved reads as done. People do the exact arithmetic the helper did. The cost of piling on context is paid in attention, and attention is the budget you are actually managing, whether you meant to or not.
The counterintuitive move is to show less on purpose. Hide the resolved tickets. Fold up the threads older than a week. Set the default view to the things that genuinely need a reply right now, and make the full history one deliberate click away instead of always in your face. A person who opens their day and sees five open items will close five. The same person staring at a wall of two hundred will close none and feel exhausted doing it. The same is true of sales pipelines, ops queues, and any tool where a human, or a helper, has to decide whether this one thing deserves action today. We learned this the hard way and it shapes how we build.
Generosity with context looks like care. Usually it is the opposite. The thing that gets read is the thing that gets answered. Everything else is wallpaper, and wallpaper quietly trains the reader, person or machine, to stop reading at all.