Everyone talks about how AI speeds things up. Running out of Claude credits and a failed ChatGPT draft both taught me how AI slowdown might be your best checkpoint for better work.
Love this framing, Katrina. The idea that friction is a feature really resonates. I’ve noticed the same thing: when AI makes something feel too easy, I sometimes charge ahead without asking if it’s actually worth doing. Those pauses you describe often create the space where the deeper insight shows up.
I especially like your point that the real skill isn’t speed but judgment. That’s such a refreshing counterbalance to the “10x productivity” narrative..
Curious. Do you have a personal ritual for leaning into those slowdown moments (like a walk, journaling, etc.), or do you just let them happen naturally?
This is 100% correct. AI isn’t a magic solution to every problem, it’s a tool. And like all tools, the output depends on both the skill and the intent of the person using it.
With an electric drill, I can put ten holes in a piece of wood by the time you can do one with a hand turned tool but ultimately, a hole is a hole.
We can use AI in the same way, and literally produce ten times the output of a person that doesn’t use it. But unlike the drill, not all of that output is created equally. Some of it is better than others.
The resultant quality of what is produced by the AI may be the same as the human; assuming the AI was prompted lazily, and the human was neither expert in the task they were given, nor were they given a quality instruction.
So in that scenario, all we have done is accelerate the production of slop. Not all AI is slop. Not all slop is AI.
Run the same experiment again; with an unskilled worker using the AI, but a skilled worker performing the task manually.
The AI will still outpace the human by 10x, but the human output will be 10x the quality.
Do it again, but this time allow the AI to be used by the human expert (assume they know how to use the tool). Now you can start to unlock productivity gains. You can increase the speed of the output and you can maintain the quality.
This is one view. Unfortunately a lot of what we’re currently seeing is a race to replace, not augment, genuine human skills with unskilled workers using AI. Quantity is maintained, quality nosedives.
Companies that pair human experts with AI will win out in the end. They can still choose to lay off and maintain output and quality with 10x fewer people, or they might choose to keep staff levels and produce 10x more output at good quality levels. Let’s see.
This isn’t to say everyone should stay in their lane and only use AI to compliment skills they have. I use it to augment genuine skills, and I also use it to provide access to skills I don’t have.
All of my Substack article images are GPT. I can’t draw, and I can’t afford (or justify the cost) of a human artist to illustrate my hobby. The human artist would do a much better job, but in this case, good enough is good enough.
Love this. Everyone talks about AI as acceleration, but the real gift is when it makes you pause. The best questions show up in the slowdown: Is this worth building? What problem am I actually solving? Sometimes friction is the feature.
Love this framing, Katrina. The idea that friction is a feature really resonates. I’ve noticed the same thing: when AI makes something feel too easy, I sometimes charge ahead without asking if it’s actually worth doing. Those pauses you describe often create the space where the deeper insight shows up.
I especially like your point that the real skill isn’t speed but judgment. That’s such a refreshing counterbalance to the “10x productivity” narrative..
Curious. Do you have a personal ritual for leaning into those slowdown moments (like a walk, journaling, etc.), or do you just let them happen naturally?
With ya. Most of the time, I don't try to complete things faster using AI. I try to do better in roughly the same amount of time.
The world is already awash in mediocre-or-worse.
This is 100% correct. AI isn’t a magic solution to every problem, it’s a tool. And like all tools, the output depends on both the skill and the intent of the person using it.
With an electric drill, I can put ten holes in a piece of wood by the time you can do one with a hand turned tool but ultimately, a hole is a hole.
We can use AI in the same way, and literally produce ten times the output of a person that doesn’t use it. But unlike the drill, not all of that output is created equally. Some of it is better than others.
The resultant quality of what is produced by the AI may be the same as the human; assuming the AI was prompted lazily, and the human was neither expert in the task they were given, nor were they given a quality instruction.
So in that scenario, all we have done is accelerate the production of slop. Not all AI is slop. Not all slop is AI.
Run the same experiment again; with an unskilled worker using the AI, but a skilled worker performing the task manually.
The AI will still outpace the human by 10x, but the human output will be 10x the quality.
Do it again, but this time allow the AI to be used by the human expert (assume they know how to use the tool). Now you can start to unlock productivity gains. You can increase the speed of the output and you can maintain the quality.
This is one view. Unfortunately a lot of what we’re currently seeing is a race to replace, not augment, genuine human skills with unskilled workers using AI. Quantity is maintained, quality nosedives.
Companies that pair human experts with AI will win out in the end. They can still choose to lay off and maintain output and quality with 10x fewer people, or they might choose to keep staff levels and produce 10x more output at good quality levels. Let’s see.
This isn’t to say everyone should stay in their lane and only use AI to compliment skills they have. I use it to augment genuine skills, and I also use it to provide access to skills I don’t have.
All of my Substack article images are GPT. I can’t draw, and I can’t afford (or justify the cost) of a human artist to illustrate my hobby. The human artist would do a much better job, but in this case, good enough is good enough.
Love this. Everyone talks about AI as acceleration, but the real gift is when it makes you pause. The best questions show up in the slowdown: Is this worth building? What problem am I actually solving? Sometimes friction is the feature.