Ask an operations manager what they do and you'll get a job description. Ask them what happened yesterday and you'll get a list of interruptions. The gap between those two answers is where most of the difficulty in coordination work actually lives.
The scheduled work — the rota, the report, the review meeting — is the part that has a shape. It appears in a calendar, it has a start and an end, and when it's done somebody notices. The unscheduled work has none of those properties. It arrives as a sentence in a message, it takes four minutes, and its only trace is that somebody is now expecting something from you.
Why it's structurally invisible
Three things have to be true for a piece of work to be manageable: it has to be captured, it has to have an owner, and it has to have a deadline. Unscheduled operational work usually fails all three at the moment it arrives.
Consider a message that says: "Hey — before Friday can you check whether the new starter's DBS came back? Sarah asked twice." That single sentence contains a deadline (Friday), a dependency (a third party who may not have replied), an implicit second obligation (tell Sarah), and a piece of context that will matter later (Sarah has already asked twice, so this is now a relationship issue as well as an admin one).
None of that is captured anywhere. To make it manageable you'd have to open a task app, write a title that loses most of the nuance, set a date, and possibly create a second task for the reply to Sarah. That's maybe ninety seconds of work to make four minutes of work trackable — and you will be interrupted again before you finish it.
The economics of capture are the whole problem. When recording the work costs a meaningful fraction of doing the work, people stop recording it.
The head is a bad database
So it goes in your head, and your head is genuinely good at this for about a day. It holds urgency well. It holds recency well. What it does not hold is the small commitment made in passing on a Tuesday three weeks ago, which is precisely the category of thing that damages trust when it's dropped.
There's an asymmetry worth naming here. Missing an urgent task is usually visible and recoverable — someone chases, you apologise, you do it. Missing a small promise is invisible to you and highly visible to the person you made it to. You don't experience the failure at all. They experience it as a pattern.
Why more discipline isn't the answer
The standard advice is a capture habit: write everything down, process the inbox daily, trust the system. It works, and it works best for people whose interruptions arrive in one channel at a predictable rate.
Operations work is the opposite of that. It arrives in a shared inbox, a personal inbox, a messaging app, a phone call, a corridor conversation, a form somebody scanned and sent as a photograph, and a forwarded thread with forty replies where the actual ask is in the middle. Asking someone to manually normalise eleven input formats into one task list is asking them to do data entry as a second full-time job.
This is the point where most tools give up and offer integrations. Connect your mail, connect your chat, and the tool will show you everything in one place. That solves aggregation and not extraction: you now have all eleven channels in one window, still unprocessed, and the job of deciding what each message obligates you to do is exactly where it was.
What would actually help
The useful move is to attack the cost of capture rather than the discipline of capture. If the thing you send in can be a voice note, a photograph, a forwarded message or eight typed words — and something else does the work of turning it into an action with an owner and a date — the economics invert. Capturing becomes cheaper than remembering.
That's the specific job we think language models are genuinely good at, and it's narrower than the way they're usually sold. Not doing the work. Not deciding what matters. Reading a messy human sentence and proposing that it probably means action: check DBS status for new starter; due Friday; waiting on external provider; tell Sarah when resolved — and then stopping, and showing you, and letting you say yes.
The proposal can be wrong. That's fine, and it's why the confidence score and the approval step exist. What matters is that the sentence stopped being something only you were holding.