The risk was on page seventeen and the summary did not think it mattered

I have had a version of this conversation more than once now, and it always arrives the same way, cheerfully.

Someone senior explains that they have not read a full document in months. Board papers, market analyses, performance reviews, all of it summarised down from twenty pages to two paragraphs. They are pleased about it, and they should be, because the time saved is real and they have plenty of other things to do.

Then you ask about something specific that was in the original. Not a trick question, just a detail that mattered, a risk buried three quarters of the way down where risks tend to live because nobody puts them at the top. And there is a pause, and the pleasure goes out of the conversation, because the summary did not think it was important enough to keep.

The pause is the thing I want to talk about. Everything else is commentary.

Not everything that gets removed is noise#

Summarising is a judgement about what matters, performed by something that does not know what matters to you.

It knows what is prominent, what is repeated, what is structurally emphasised. It does not know that you have been circling a particular supplier relationship for eight months, or that the sentence on page seventeen connects to something you were told in a corridor in March. Those connections are exactly where the value in reading actually lives, and they are invisible to anything working from the document alone.

The parallel with food is not decoration, it is close to exact. Processing removes what does not survive processing well, which is usually the fragile stuff, and adds a smoothness that reads as completeness. A summary and a document feel like the same thing consumed at different speeds. They are not. One has the warnings, the qualifications, the sentence where the author's confidence visibly wobbles. The other has the conclusions, tidied.

You can live on this quite happily for a long time. The bill arrives later and is not itemised.

Friction is doing something#

There is a phrase I keep returning to for the thing being removed, which is productive friction. The bits of work that feel wasteful and are not.

Reading the whole thing so that a detail lodges somewhere and surfaces two months later. Drafting the argument badly first, which is how you discover it does not hold. Arguing with a conclusion until you can articulate why you disagree. None of it is efficient, all of it is where understanding is manufactured, and every one of these is available for automation right now.

The stairs analogy is obvious and I am going to use it anyway, because it is exactly right. The lift is better in every measurable respect. It is faster, it requires less effort, and if you take it every day for two years something has happened to your legs that no single day caused.

The gap that nobody is measuring#

Here is the part I find genuinely uncomfortable, and it is not the part usually written about.

People who were curious before all this are using these tools to go considerably deeper than they previously could. They interrogate, they push back, they use the machine as something to argue with, and their work has visibly improved. That is real and I am one of them on a good day.

People who were content with the surface now have a tool that makes the surface glossier. They were never going to read page seventeen. The difference is that previously their shallowness was visible, in the vagueness of what they produced, and now it arrives beautifully formatted and reads as competent.

So the distance between the two groups is widening quietly, and none of the usual measures catch it, because both groups show identical adoption statistics and comparable output volume. One group is compounding capability. The other is compounding presentation. And the second group is not lazy or foolish, they are responding rationally to a system that has never once rewarded them for reading the whole thing.

Where I stop weighing it up#

I try to hold most of these questions open, because the honest position on most technology is that it trades one thing for another and reasonable people weigh the trade differently.

This is the place I stop doing that. Some of what is being handed over is not ours to trade. Institutional memory belongs partly to the people who come after us. The capacity to think a problem through unaided is not a personal preference, it is the thing that lets an organisation notice when its systems are wrong, and once it is gone there is no external alarm that replaces it.

I am not arguing for less automation. I am arguing that this particular cost should appear somewhere in the calculation, and at present it appears nowhere at all, because it is slow, diffuse, and shows up in a different budget year from the saving.

Three things that actually work#

I am wary of ending on concern without equipment, so here is what I have seen make a difference, none of it dramatic.

Spot-check summaries against sources, unpredictably and regularly. Not everything, and not on a schedule people can anticipate. The value is partly in what you find and mostly in what the practice does to how carefully people read what they receive.

Build in a pause between receiving output and acting on it. Not a formal review stage, which becomes a rubber stamp within a quarter, just an interval. Instant action on instant output is how errors travel fastest.

And make challenging the machine somebody's actual job, rotated, so it does not become one person's tiresome reputation. The person who finds the flaw should be visibly valued rather than treated as an obstacle to throughput, because whether that role is rewarded or resented determines everything about whether it survives.

The question I cannot answer#

Judgement is built by being present when things go wrong. That is the only way anyone has ever acquired it.

If the work that used to build it is going to the machines, and the graduate never does the tedious reconciliation that teaches them what a wrong number feels like, where does the next generation's instinct come from?

I do not know. I have asked a lot of people and I have not heard an answer that convinces me, including my own. It is worth sitting with rather than resolving, and I would rather leave it open on the page than close it neatly and be wrong.