AI made my humanities degree feel technical

For years, I thought of my English degree and journalism training as useful in a fairly traditional way.

They taught me how to read, write, interview, structure an argument, question a source and work out what somebody was actually trying to say. Those things helped me become a journalist and later a content writer.

But I never really thought of them as technical skills.

If anything, I tended to put them on the other side of the line. Technology belonged to people who could code, configure systems, understand databases or build software. I worked with words.

AI has made that distinction feel much less stable.

The technology has not made my degree more technical. What has changed is the interface.

Natural language has become one of the ways we operate increasingly capable computing systems, and suddenly a lot of the things I spent years learning how to do with language have much more direct technological leverage.

I am starting to wonder whether Editorial Intelligence itself sits partly inside that shift.

The machine moved towards my strongest medium

Most software I have used in my career asked me to learn its interface.

You learned where the buttons were. You learned the menus, shortcuts, fields, query syntax or workflow somebody else had designed. More technical systems required more specialist knowledge.

Large language models are different.

The interface is conversation.

That does not make the underlying technology simple. Far from it. But it changes what a non-engineer can do with it.

I can describe a half-formed idea, ask the system to test it, introduce new evidence, reject the first framing, point out a contradiction, ask for a different structure and keep moving until the argument becomes clearer.

That feels natural to me in a way many earlier forms of computing did not.

It also feels suspiciously close to editing.

The useful skill is not knowing a magic prompt. It is being able to say: that is nearly right, but you have misunderstood the distinction I am making. Or: the evidence does not support that sentence. Or: you have made this cleaner by removing the part that actually matters.

Those are language instructions, but they are also judgement.

For the first time in my career, the technological interface has moved significantly towards the medium I have spent most of my working life learning to control.

English taught me to live with ambiguity

An English degree did not teach me how to use AI.

It did train me to do something that turns out to be useful when working with it: stay inside ambiguity long enough to work out what a piece of language is actually doing.

You read a text and ask why one interpretation feels stronger than another. You notice tone, structure, omission, contradiction and the difference between what is stated and what is implied. You learn that two plausible readings can coexist and that language can sound confident while remaining unstable underneath.

That is useful around generative AI because its outputs are often plausible before they are necessarily right.

A fluent answer can conceal a weak distinction. A neat summary can remove uncertainty that should have survived. A rewritten paragraph can be cleaner while becoming less true.

Close reading suddenly has an operational use.

You are not simply consuming the output. You are interrogating it.

Journalism added the evidence layer

Journalism added a different set of instincts: Who is saying this? How do they know? What is missing? What would the other person say? Is that actually evidence, or just a confident claim? What question should I ask next? What changed between the version I expected to hear and what the source actually told me?

Those habits matter with AI because language models are extremely good at producing an answer-shaped object. That does not mean the answer deserves to survive.

I have written quite a lot recently about the importance of evidence, opposition and editorial judgement in AI workflows. In Writers learn what not to write, I argued that experienced writers carry a large amount of negative knowledge: patterns they distrust, claims they know need checking and structures they have learned not to accept at face value.

That experience becomes part of the interaction with the model.

I am not approaching a draft as a blank consumer asking whether it sounds good. I am approaching it with years of accumulated reasons to stop, question, cut, challenge or ask again.

That may be one reason AI has felt less like learning a completely new craft and more like discovering a new surface for an old one.

This is more than prompting

I think “prompting” sometimes makes this look shallower than it is. The phrase can suggest that the valuable skill is learning how to ask the machine for the right thing. Sometimes it is, but a lot of my own AI use does not feel like that anymore.

It feels like keeping a complicated thought alive across a sequence of interactions. The first answer is rarely the point. It gives me something to react to. I notice what is wrong, introduce context, narrow the argument, reject a false simplification and decide which part is worth preserving.

The important ability is not communicating at the AI.

It is being able to externalise thought precisely enough that another system can participate in it without quietly taking ownership of the judgement.

That connects to my argument that prompting is becoming routing. As the systems become more capable, the work shifts upwards. Which tool should handle this? What context does it need? What evidence should it see? What belongs in durable memory? Where should a second perspective challenge the first? What needs a human decision?

Those are system questions.

But many of the decisions inside them are still deeply interpretive.

Maybe Editorial Intelligence is partly a humanities discipline

This is where the idea becomes more interesting to me.

I have been describing Editorial Intelligence as a way of preserving and applying editorial judgement across AI, knowledge systems, evidence and workflows.

That can sound technical because some of the implementation is technical. There are repositories, structured fields, routing decisions, memory layers, retrieval, agents and automation.

But what are those systems trying to preserve?

• Context and meaning
• Evidence and narrative
• Decisions, contradictions and changes of mind
• The reason one claim survived and another did not
• The judgement required to decide whether two pieces of information genuinely belong together

Those are not exclusively humanities concerns, obviously. But they are familiar ones.

Perhaps part of Editorial Intelligence is the attempt to make traditionally humanistic capabilities operational inside computational systems:

• not just “write this in the right tone”, but preserve why this argument exists
• not just retrieve a source, but understand what the source can actually support
• not just connect two documents, but decide whether the relationship between them is meaningful
• not just generate options, but interpret which option belongs in this context
• not just remember a decision, but keep enough of the reasoning around it that somebody can challenge it later

That is much closer to interpretation than automation.

I still would not call humanities a substitute for technical expertise

There is an obvious danger in getting carried away with this.

Being good with language does not make somebody an engineer. An English degree does not teach model architecture, statistics, APIs, databases, software design, security or any of the other disciplines required to build the systems I am using.

Natural-language interfaces can also create false confidence. It is now possible to make something technical happen without fully understanding what happened underneath. That can be liberating. It can also be dangerous.

So I do not think the lesson is that humanities graduates have suddenly become technologists, or that technical expertise matters less.

The more interesting change is that the boundary has become more porous.

Technical systems can now accept much richer instructions from people whose strongest capabilities sit in language, interpretation and domain knowledge. Those people can do more with technology without pretending they built the technology.

That feels like a healthier claim.

Maybe this is why “content engineer” feels both right and wrong

I recently asked Am I a content engineer?.

My answer was deliberately unresolved. Some of my work increasingly involves designing systems around content: evidence banks, editorial memory, AI workflows, structured customer intelligence and routes between different tools.

That looks more engineered than the content work I was doing ten years ago. But perhaps the part I was missing is that the system is being built around a very old set of capabilities:

• read closely
• ask better questions
• understand what somebody means
• challenge the source
• notice what is absent
• distinguish evidence from assertion
• keep contradictory possibilities alive until the evidence resolves them
• choose what matters and explain it clearly

Those skills did not suddenly become technical. Technology became much more capable of responding to them. That distinction matters to me because it changes how I understand my own route into AI.

I did not leave writing and journalism behind to become fluent with these systems. I brought those disciplines with me.

And perhaps Editorial Intelligence is my attempt to work out what happens when the habits of the humanities meet machines that can finally work through the same interface. For years, I assumed my humanities background sat somewhere adjacent to technology. AI has made me wonder whether language, interpretation and judgement are becoming part of the technical surface itself.

The degree did not change. The interface did.