I have a confession to make about organisational performance art. First, I need to take you back to the late nineties.
At university, I was not a devoted library person — even though this was a time when it was really the only place to find information. I could admire the idea of stacks of books, colour-coded notes and the faint smell of panic around a photocopier, but I was more likely to swerve it, work out what the question was really asking, and then bang out an essay with a really strong structure, a clear argument and enough confident prose to make it stick. It freed me up to finish early and save a table at the pub.
More often than it probably should have, this approach worked very well for me.
One way of looking at this is that I dealt with my regret at taking a Classics module by dodging serious work and tricking my lecturer into giving me better grades than I deserved, because I had realised he liked a nicely framed paragraph. There is also a more respectable version of that lesson: structure matters, and clarity matters. If you are asking someone busy to understand a complicated issue, make a decision or approve something, it is not a mark of intellectual seriousness to make them excavate the key point from six paragraphs of background and 74 footnotes. Structure and good writing have real value.
I should probably not say that some version of this habit followed me into professional life as a civil servant. But anyone who has written a briefing, paper or presentation knows that the craft matters. If anything, we get more honest about that as we get older. A good briefing tells someone what they need to know, what they are being asked to decide and why it matters. It is not “style over substance”; it is a service to the reader.
But there is a limit
My film studies lecturer, on the other hand, was very good. He would know for damn sure if I had not actually grasped the nuance of early Soviet cinema. I could write something elegant, make an argument based on a half-remembered seminar, and deploy the word “montage” with enough confidence to get through the first paragraph. But if I had not understood what the film was doing — its history, politics, form, technique and the particular force of a scene — he would know. On top of that, I was enjoying film studies and didn’t mind watching films and reading about them. I was happy to do the work and write it up nicely.
The point is: you should know when you can’t hide behind a well-turned sentence.
I have been thinking about this since generative AI properly arrived in the workplace. AI has made it possible to produce the appearance of organisational competence, across more people, at quite extraordinary speed.
Reassuring artefacts
Organisations have always produced the things that mainly exist to show they are being properly run: meeting papers, steering group packs, executive summaries, risk registers, action logs, minutes, programme updates and emails copied to everyone who might conceivably need to know. Always adding one more person for luck.
Some of this is essential. Paperwork can make a risk visible and mean that something is actually done about it. It can even stop the person with the loudest voice — or the best informal access — being the only person who knows what is going on.
But there is also the other stuff — the stuff that has its origin story in essays like the ones I handed my poor old Classics lecturer. It is the meeting paper produced because there is a meeting in the diary. The email copied to half the organisation because nobody wants to be the person who, three weeks later, is accused of failing to keep someone in the loop.
Meanwhile, a project team somewhere in an open-plan office will do the same thing they would have done if none of that had happened.
When the production line speeds up
AI has not created this problem, but it has made the production line much faster. A rough set of notes can now become formal minutes; the minutes become an action log; the action log becomes a status update.
Each individual step can seem entirely reasonable. A meeting note is useful, and different audiences do need different information. The problem is that AI makes it very easy to be overly diligent; you can add a bit more context and include a few more caveats. At some point, you have created an organisation in which everyone has received everything, and nobody has read it all.
Microsoft’s 2025 analysis of workplace activity found that the average worker received 117 emails and 153 Teams messages on a weekday. We can debate the numbers, and Microsoft is not exactly a neutral observer of workplace technology, but the experience is familiar enough. The issue is not that people cannot produce words quickly enough to feed the beast; it is that other people’s attention is finite. If you are mostly in meetings, how many of those emails are you really going to read?
Sometimes, performance is the job
This is not, however, an argument for stripping organisations back to a single cryptic email and a vague verbal agreement. Sometimes it is completely reasonable to produce the two-page version for the big boss when your immediate boss needs the longer paper, the detail, the evidence and the citations. The senior audience may need the decision, the implication and the thing that could go wrong. The person closer to the work may need to see how you got there, what the assumptions are, what has been discounted and whether the conclusion stands up. They could, in theory, just take your word for it, but the extra version can be a crucial part of doing the job properly. Using AI to make it more quickly is legit.
Sometimes a polished presentation creates confidence because people in the room need to know that someone has thought about the detail, and a well-managed meeting creates conditions for disagreement that wouldn’t happen over email. A bit of performance is not necessarily dishonest: it can be reassurance, translation, courtesy, access and accountability.
The trouble starts when we stop being selective about it — when every audience gets the full set of materials because it is easy to create the content, or when every meeting generates a paper because that is what meetings do.
Hollywood or Soviet montage?
There is a proper film studies way of explaining this, and I am aware that I am about to make a distinction that could cause somebody to throw a small book at me. Or an old VHS copy of Battleship Potemkin.
Broadly the argument works like this: a lot of classic Hollywood filmmaking is specifically designed to make you accept the world it has created without getting distracted by having to think too much. The editing is smooth, the logic is clear, the story carries you along. You do not spend the whole time noticing the joins. You are meant to feel that this is simply what happened. And it works.
A lot of Soviet cinema — the kind you have to write essays about — does something different. It wants you to notice the joins. The films put images alongside each other in ways that make you think. The directors want you to see the tension, make the abstract connection, spot the politics or wonder about how it was all constructed in the edit.
Neither approach is necessarily better; not every situation calls for the meeting note version of difficult Russian art. Sometimes the point of communication is to make a complicated world manageable enough for somebody to act. Other times, the point is to stop them accepting a neat story too quickly.
AI can do both. It can smooth things out, but it can also help show the joins: identify assumptions, flag missing evidence, compare the reassuring narrative with the data, or ask what a sceptical finance director — or difficult customer — might say. The skill is knowing which job we need it to do: smooth the joins so people can act, or expose them so they can think.
And that, readers, is the longest answer I have ever given to the question of why Leading AI’s tools provide citations as standard. Clarity matters — but so does being able to see how you got there.
