AI-assisted Craft: A Manifesto
The missing category between human-made and AI-made.
The AI slop debate has been circling the same drain since 2024. By now, you’d think we might have found a name for the position somewhere between “artistic crime” and “the machine as the ultimate collaboration partner.” We haven’t.
One side calls every use of AI creative fraud. That side is especially active on Facebook, where Meta’s algorithm happily reinforces the belief. The other presses “Generate” and calls it collaboration. A life-changing, millionaire-making meeting of minds. One of those minds typed six words and the other predicted the rest, but let’s not ruin the hype.
Meanwhile, many of us are over here using AI to do solid work.
There’s a way to create with AI without watering down your claim to the work. And that practice needs a name.
AI-assisted craft.
What Is AI-Assisted Craft?
Before attempting a small intellectual land grab of naming this type of work, I searched for the phrase, fully expecting someone to have used it before me. They had. But in a more narrow context and without a definition sturdy enough to be useful.
So here is mine.
AI-assisted craft means the human sets the intention and the standard for the work, then directs how it gets made. AI gets a defined supporting role. Supporting, as in: it does not get to make the decisions. The practice applies across knowledge work and digital creation, from writing and research to coding and design.
Look. LLMs for research, coding, learning, challenging your thinking, or drafting aren’t radioactive contraband. They’re tools. Use them.
Hey, I’m Karo Zieminski 🤗
AI PM and builder. I write Product with Attitude, an AI newsletter for tens of thousands of readers across 146 countries, helping them develop critical AI literacy the only way it sticks: through practice.
If you’re new here, welcome! Here’s what you might have missed:
How Is AI-Assisted Craft Different From AI Slop?
The same way hand-painted Chinese porcelain is different from a plastic cup on Temu: by the amount of care and skill that went into making it.
Slop begins where human care ends, regardless of whether AI was involved. Indifference, not AI, is the breeding ground for slop. It happens when the maker doesn’t care enough about the human audience to offer them quality.
I first encountered the term through Simon Willison in 2024. His definition was brutally efficient: unwanted AI-generated content.
Back in 2022, when AI slop looked deranged, it was easy to spot. Twelve-fingered toddlers stared from portraits. By 2025, slop had learned to code and ship apps in purple gradients. Now, in 2026, it comes with an AIO strategy: pages answering FAQs no human has ever asked, but the bots might. I bet that by 2028, it will have weaponized personalization. It will arrive custom-made for you.
If you’re reading this newsletter, you can probably spot slop from a mile away. Can your loved ones?
The Two Axes of AI-Assisted Craft
Axis 1 asks who produced the primary substance: human-made, AI-assisted, AI-directed, AI-generated, or AI-autonomous. It doesn’t grade the result.
Axis 2 asks how much craft the work shows: deliberate, competent, negligent, or slop. This describes quality. It doesn’t tell us which tools were used.
Production labels (Axis 1)
Human-made. A person produces the work. The choices, mistakes, revisions, and final form are theirs. The label doesn’t tell us which tools they used.
AI-assisted. The human is the primary maker. The choices, mistakes, revisions, and final form are theirs. AI is one of the tools they use.
AI-directed. The model produces the work; the human directs it. That direction must be consequential enough to shape the result. One prompt followed by a shrug is generation with supervision theatre.
AI-generated. AI-generated content is a neutral term. It simply means that the model produced the primary substance of the work. It says nothing about quality. Nano Banana images are AI-generated content. They can be stunning, or they can prove that photorealism and taste aren’t the same thing. The label only tells us how the pixels arrived.
AI-autonomous. You don’t work on tasks, and instead design the conditions under which the work is produced. NIST AI Risk Management and Human-AI Interaction describes a similar spectrum, from fully manual to fully autonomous. There can be plenty of craft in that setup. There can also be a tangle of defaults that nobody checked because the system seemed to be working. “Autonomous” doesn’t settle the question.
Quality labels (Axis 2)
Deliberate. Planned and executed with care. Nothing is there by accident.
Competent. Correct, solid, unremarkable. It does the job.
Negligent. Someone should have checked and didn’t.
Slop. AI or no AI, the work lacks of quality, judgment, and care.
How Can You Tell if the Work Shows Craft?
I wanted to design a test that works equally well for an article, a product specification, a piece of code, a research report, or a visual design. So the questions are deliberately tool-agnostic.
The prompt test: This one is my favorite. Can I guess, just by looking at the work, what prompt could have produced it almost exactly as it appears? If I can, it isn’t personal or specific enough. More AI than craft.
Facial expression test: Can I tell who the work is for? Or is it hotel-lobby music trying to satisfy the entire internet? Does it avoid saying anything that might cause a facial expression? AI is brilliant at measured opinions. Humans, less so.
Verification: Are all the claims factually correct? Basics, really.
Presence test: Did I learn anything new about you?
One failed check doesn’t automatically make the work slop. Several failures, however...
The 6 Rules of AI-Assisted Craft
This manifesto is deliberately small. Six rules are enough to test any workflow.
Bring Your Own Opinion (BYOO)
Don’t ask Claude what you believe. Start with something that belongs to you: an opinion, a question, an observation, or a problem that refuses to leave you alone. It can be scruffy. Mine usually is.
AI can help you wrestle it into shape. A perfectly adequate job for a machine.
I have one skill that turns unfinished thoughts into a research plan and another that does the same for my dictated Wispr notes. Lovely tools. Neither gets to decide what I mean.
Let AI organize the mess. But bring the opinion yourself. We'll know if you don't.
Give AI a Bounded Job
“Improve this” is not a bounded task.
Give the model a job with sharp edges: do this, not that.
Research for me, BUT bring me facts, not conclusions. Challenge my assumptions, BUT do that through Socratic questions so I have to do the thinking. Explain this code block, BUT test whether I understood it. Suggest design fixes BUT don't bleach my personality out of it. Whatever it is, leave space for my judgment.
In this context, a bounded AI task has three parts: a specific purpose, a BUT that limits the model’s role, and a final decision that still belongs to me, the human. When Microsoft Research surveyed 860 of its own developers, they wanted the same arrangement: for AI to absorb “the assembly work surrounding their craft, never the craft itself.”
Remember Your Person
Are you creating for someone (that someone can be you) or for $$$? We can tell.
I show you exactly how to find your person in this article.
Verify
Check everything. An invented citation can sit beside a real one without so much as twitching. Code that you ran successfully once has proved exactly one thing: it ran once.
Everything Fits. Do You?
AI has tells. Familiar turns of phrase, predictable visuals. There is nothing reaaaaally wrong with them, but are they yours? Would you have chosen this metaphor, or put that visual on your site? If the answer is no, why is it still there?
Your creativity has tells too. They echo one unique source: your lived experience. Can we see what it taught you? Or have AI’s defaults sanded every trace of it away?
Own It.
With your name and your reputation. All of it.
Keep Reading
I love how you defined “AI assisted craft.” This is exactly how I think about AI. It should be a copilot that helps us refine ideas, think through our work, and get things done, not something that replaces the human thought, creativity, and lived experience behind it.
The human touch is what gives the work meaning. AI should support that, not become a substitute for it.
Thank you Karo. This transparency and opportunity for dialogue is what is sadly missing from many of the witch hunts that are starting to emerge elsewhere in this platform. 😢