Most AI writing tools today are built on a single assumption: that what writers want is for the AI to write for them. Faster output, less effort, more text in less time. But there's another way to use AI in writing — as a partner that helps you think rather than a ghost that produces text in your place. Cognitive offloading research (Risko & Gilbert, 2016) shows the trade-off clearly: offloading a skill changes that skill over time. The choice between generation and partnership isn't really about AI. It's about what you believe writing is for.
There are two basic ways an AI tool can sit next to your writing.
The first posture is generation. You give it a prompt; it gives you text. You give it a paragraph; it gives you a polished version. You give it an outline; it gives you a chapter. The AI's job is to produce. Your job is to direct, accept, or reject what it produces. This is the dominant posture today.
The second posture is partnership. You think out loud; it asks questions. You get stuck; it surfaces what you've already half-said. You make a claim; it offers a counter-perspective. The AI's job is to help your thinking. The text remains yours. This posture is rarer, and it's the one Wretify is built around.
AI writing tools sit in one of two postures: generation (writes for you) or partnership (thinks with you). Cognitive offloading research shows that offloading a skill changes that skill over time. The choice compounds across years of use.
What's actually happening with each posture?
The generation posture is doing something we should be honest about. Every time the AI writes a sentence you accept, you've outsourced a small piece of thinking. Not because you're lazy — because that's what the tool is designed to enable. The outsourcing is the value proposition.
For some kinds of writing, this is fine. Generating a routine email, writing boilerplate text, producing a first draft of something nobody will read closely — these are tasks where the thinking doesn't matter much, and outsourcing the production is a clean win.
For writing that's for something — a book, an essay, a body of work meant to express what you actually think — the outsourcing has a different character. You're outsourcing the part where the thinking happens. The cognitive scientists Evan Risko and Sam Gilbert, in a much-cited 2016 paper for Trends in Cognitive Sciences, formalized a concept called cognitive offloading. Their argument was that offloading is rational and useful when the tool gives back what you put in. Calendars offload memory; we get better at scheduling. GPS offloads navigation; we get worse at spatial reasoning. The trade-off depends on whether the underlying skill matters.
For nonfiction writers, the underlying skill — being able to think a hard thought through to clear language — is the whole game. Offloading it produces text but not the writer who wrote it.
The partnership posture takes a different bet. It bets that the value isn't in the speed of production. It's in the quality of the thinking. The Russian psychologist Lev Vygotsky, working in the 1920s and 30s, called this kind of scaffolded help the zone of proximal development — the range of work a person can do with assistance but couldn't yet do alone. Good teachers, coaches, and editors operate in this zone. They help you do work that's slightly beyond your current reach, in a way that expands your reach over time. They don't do the work for you.
An AI partner is trying to be that kind of presence. It's not faster than a ghostwriter; it's often slower. But the writer at the end of the process is a different writer than the one at the beginning.
Why doesn't "the AI is just a tool" hold up?
The standard defense of AI writing tools is that the AI is just a tool, like a typewriter or a spell-checker, and what matters is what the human does with it. This is true in a narrow sense and misleading in a broader one.
Tools shape what they're used for. The cognitive psychologist Donald Norman, in his 1988 book The Design of Everyday Things, called this affordances — the actions a tool invites and the actions it discourages. A hammer affords hammering. A keyboard affords typing. An AI that generates polished text affords letting the AI generate polished text. You can use it for something else, but you're working against the grain.
This matters because the affordance accumulates over years of use. A writer who has spent five years generating text with AI is, on average, a different writer from one who has spent five years thinking with AI. The first has produced more words and developed a thinner relationship to their own voice. The second has produced fewer words and developed a sharper one.
Neither path is wrong. But the choice is real, and it's worth making consciously.
What actually helps?
The principle is simple: choose AI tools whose affordances match what you want writing to be in your life.
If your goal is volume — producing a lot of text quickly, for purposes where the thinking is incidental — generative tools are the right choice. There's no shame in using them, and they're often the right tool for the job.
If your goal is to develop as a thinker through writing — to use the page as a way of finding out what you actually believe — partnership tools are the right choice. They're slower, more demanding, and harder to evaluate by output count. They reward you in a different currency.
The signs of a partnership-oriented tool include: it asks questions rather than supplying answers. It surfaces what you've already said rather than generating new claims. It treats your voice and worldview as inputs to honor, not defaults to override. It makes you do the writing yourself, even when it could do it for you. It rewards depth of thinking rather than speed of production.
How Wretify approaches this
Wretify's AI brainstorming partner is built around questions, not output. You tap the microphone and say what's on your mind. The AI distills what you said and asks the question that might take you deeper. You answer. The AI asks again. Round by round, what was vague becomes clear. What was stuck starts moving.
There's no "write this paragraph for me" button. There's no "generate a chapter" feature. The AI doesn't produce the text. You produce the text. The AI helps you find what the text wants to be.
This is by design and it's the most contested design choice I made. Many writers, on first encounter, want the generation features. They want the tool to do more. Some find their way to using Wretify as a thinking partner and stay. Others go back to tools that produce text for them, and that's a legitimate choice — different tool, different relationship with writing.
I built it this way because the kind of writing I care about — and the kind most people pick up nonfiction to do — is the kind where the writing is the thinking. Skip the thinking and you've skipped the point.
The kind of writing most people pick up nonfiction to do is the kind where the writing is the thinking. Skip the thinking and you've skipped the point.
- AI writing tools sit in one of two postures: generation (writes for you) or partnership (thinks with you).
- Cognitive offloading (Risko & Gilbert, 2016): offloading skills changes those skills over time — for better or worse.
- Vygotsky's zone of proximal development: scaffolded help expands a person's reach without replacing their effort.
- Affordances (Norman, 1988): tools shape what they're used for, regardless of the user's intent.
- Wretify's AI asks questions rather than supplying text — the writing is the thinking, by deliberate design.
This article was written in Wretify. Try it free →
Frequently asked questions
No. They're the right tool for some jobs — high-volume routine writing, text where speed matters more than voice, situations where the thinking is incidental to the output. The question isn't whether they're good. It's whether they're right for your writing.
In raw output terms, often yes. In meaningful-words-per-month terms, often no. Partnership tools produce less drafting and more thinking; the writing that does emerge tends to be writing you'd stand behind.
Yes, and many writers do. The point isn't to be doctrinaire. It's to be aware of which posture you're in at any given moment, and what that posture is doing to your relationship with the work.
The difference is real and observable in the output. A writer who has been thinking with AI for a year produces writing that sounds like them, with their own contradictions and voice. A writer who has been generating with AI produces writing that sounds like fluent average prose. Both are recognizable.
That writing isn't really about producing text. It's about finding out what you think. Text is a side effect of that finding-out. Tools that skip to the side effect skip the part that matters.