Writing Things Down: Why It Still Matters in the AI Era

The Order Things Are Said Is Not the Order They Should Be Thought

At a consulting firm, one of the tasks I was drilled on relentlessly was writing up meeting notes. Not chronological transcripts of who said what, when. What was the meeting actually trying to achieve? What were the real points of contention, for what reasons did we arrive at a conclusion, what was decided, and what was left unresolved? I was expected to organize all of that in my own head and write it "structured."

"The order things were said" and "the order things should be thought" are not the same. Notes written without understanding that were pushed back at me over and over. At the time, it felt like a harsh, tedious kind of training. Looking back, though, this experience raised my business skills more than almost anything else. What was being trained was not the elegance of writing. It was the ability to read meaning out of unorganized information, define the real points, and reassemble it into a form that other people could actually make decisions on.

Writing Breaks the Illusion of Understanding

I feel the importance of this every day in current work. Engineers do not sufficiently understand the front-line operations or the business context. The business and operations side, on the other hand, does not understand the constraints of the system or the technology. This is not necessarily anyone's incompetence. Their fields of expertise and their roles are different, so to some extent this is a given.

But business value is not something each specialty produces on its own. The problems of the front line get correctly translated into the system, and the technical possibilities and constraints are fed back into business-side decisions. Value emerges only through that connection. The problem is that the ambiguity between the two sides tends to be left untouched.

When people talk out loud, they feel in the moment as though they have understood each other. In reality, they are often using the same words with different meanings, or leaving out the scope and edge cases. Sometimes the parts they cannot really explain get covered up with jargon, and the conversation just moves on. Once you write it down, you have no choice but to spell out the subject, the conditions, the causes, the impacts, and the criteria for judgment.

What matters in doing this is to line up as little jargon as possible and to write in plain language that a third party seeing it for the first time can follow. What is happening, why is it happening, whom does it affect and how, and what has to be decided. Language that lets someone follow the chain of cause and effect without any specialist knowledge. This is not mere rewording. It is work that the writer themselves cannot do without having genuinely chewed on and understood the content.

Of course, this doesn't mean removing all specialist terms. In the details required for implementation and operations, precise technical terminology and data definitions are necessary. But upstream of that, you need a shared language in which someone outside the specialty can still understand what is happening and what the decision means. The point is not to fully share the specialist knowledge — it is to share the causal relationships that cross specialty boundaries.

AI Can Take Over Not Just Writing, but Thinking Itself

With the arrival of generative AI, the difficulty of this work has dropped dramatically. AI can pull the real points out of a transcript, translate technical explanations into plain language, and even flag which conditions have been left unspecified. As a tool for helping the business side and the technical side deepen mutual understanding, AI is extremely useful.

At the same time, AI can also shape fuzzy understanding into fluent text where the fuzziness is no longer visible. Polished writing and correct understanding are not the same thing. What is the real problem? How much of this is fact and where does inference begin? Which information should we keep, and which should we drop? Hand judgments like these over to AI as well, and humans end up giving up not only the writing, but the initiative of thought itself.

"AI can write, so you don't need to write yourself" is a defensible view. It is true that the necessity of writing everything from scratch, every time, purely by hand, has decreased. But to catch the gaps and the logical leaps in what AI produces, you have to be able to structure text yourself.

Baseline Strength for the Young, Amplifier for the Veteran

That is exactly why I hope young people who are just starting to build their work capability will actually practice writing. Take the unorganized information of meetings and the field, understand it yourself, give it structure, and turn it into text that other people can act on. This work trains the underlying muscles common to every kind of job — defining the real points, tracing cause and effect, deciding what to include and what to leave out, taking the reader's perspective, being accountable for what you say. To use a sports analogy, it is like building the inner muscles that support the body, before learning the flashier techniques.

At the same time, this is not training only for the young. It is also needed by those who have already become experts in a specific domain, and by veterans with long experience. The more experience you accumulate, the wider the circle of people with whom you can communicate purely in jargon and unstated assumptions — and the harder it becomes to notice what you are leaving out. Can you get that specialist knowledge out in a form that a third party seeing it for the first time can understand? Can you convert it into something that connects with another specialty and leads to a real judgment or execution? That is where experience stops being a personal stock of knowledge and becomes value for the organization.

The same applies to using AI. AI can organize, compare, expand, and verify at high speed based on the information you give it. But if the premises, the goals, the real points, and the causal relationships are left ambiguous, all you get back is plausible-sounding generalities. If experts and veterans can put their knowledge and judgment into text that is plain and structured, AI becomes more than a text machine — it becomes a tool for extending knowledge, connecting different domains, and generating new options. The more precise the writing, the more precise the use of AI, and the greater the added value that emerges from it.

What should sit at the center of the training is not writing in volume. It is understanding unorganized information, deciding on a structure, writing it out, receiving other people's pushback, and rewriting once more. It is repeating that process.

The writing skill demanded in the AI era is not the ability to rapidly produce elegant text. It is the ability to find the ambiguity between specialties, structure it in plain language, and create a state in which people from different positions can make decisions based on the same facts. Writing is not just a means of transmission. For the young, it is a way to build the baseline strength of work itself; for experts and veterans, it is a technique for connecting accumulated knowledge with others and with AI, and turning it into greater value.

← Back to Thoughts