I'm building an AI communication tool, so I'm going to do something slightly uncomfortable and show you how I actually communicate with AI.
No spellcheck. No cleanup. No pretending this was a carefully engineered prompt.
Recently, while working on an update to my website, I wrote something along these lines:
“you are my senior graphic and web deisgner.. help me with a quick wensite update… position myself as beign more marketing focused… influencer agency… testing teh waters…”
And later:
“create aprompt for me to kill this update in a signle gol without putting strain on my agentic costs… as close to free as possible.”
Yes. Those are real typos. And I'm leaving them there intentionally.
Because here's the more interesting question:
You understood what I meant, didn't you?
There is actually a surprising amount of information packed into those badly constructed sentences.
I established the role I wanted AI to play. I explained that I was updating an existing website, not rebuilding it. I wanted my positioning to lean more heavily into marketing. I introduced Vibe Agency, an influencer talent agency I've been developing. And “testing the waters” communicated something important too: this is a new venture, so don't suddenly reposition my entire professional identity around it.
The second prompt added another set of constraints. I wanted the job done in one pass. I didn't want endless iterations. I wanted to minimize agentic costs. And I wanted AI to take more responsibility for figuring out how to accomplish that.
The writing was messy.
The intent wasn't.
We may be solving the wrong AI problem
I've spent a lot of time thinking about what it actually means for AI to “sound human.” And I think we're solving the wrong problem.
We keep trying to make AI sound more like us. Make it conversational. Give it personality. Vary the sentence length. Use contractions. Stop using certain words. Stop using em dashes. Add imperfections so nobody realizes a machine was involved.
Basically, we're getting better at teaching machines to impersonate people.
But human communication isn't defined by contractions, sentence length or strategically placed imperfections. It's defined by the person communicating. And every one of us does that differently.
Some people organize their thoughts before they speak. Others discover what they think while they're explaining it. Some people are extremely direct. Others give you six pieces of context before getting to the point because, to them, the context is part of the point.
Some people communicate in perfectly structured sentences. And some of us apparently type “signle gol” while simultaneously worrying about agentic computing costs.
That's human communication too.
The person can disappear in the rewrite
Give AI a messy paragraph and ask it to: “Say this better.”
Watch what happens. The grammar improves. The sentences get cleaner. The ideas become organized. The unnecessary words disappear.
Technically, it worked. Except sometimes something else disappears too.
You.
We've gotten very good at using AI to improve writing. I'm not convinced we've gotten equally good at using it to improve communication.
Those aren't the same thing. Writing is the output. Communication is the transfer of meaning from one human being to another.
If I send my best friend a message, I communicate differently than I would with my boss. If I'm writing on LinkedIn, I communicate differently than I would during a presentation. If I'm explaining something I'm excited about, I probably communicate differently than when I'm trying to resolve a disagreement.
I'm still the same person. What changes is the transmission.
Preserve the brain. Improve the transmission.
The brain is the person behind the message. Their ideas. Reasoning. Personality. Intent. Humor. Context. Even the strange ways they connect one thought to another.
The transmission is how those thoughts need to reach somebody else. That's where I want AI helping me.
Maybe I buried my strongest point halfway through a paragraph. Move it.
Maybe I gave someone context they don't need. Trim it.
Maybe something sounds defensive when I don't intend it that way. Fix it.
Maybe I'm communicating with an executive who has 45 seconds to understand something I spent three hours thinking about. Help me translate the complexity without stripping away the idea.
And yes, if I misspelled “website,” please fix that too.
I don't want AI preserving my mistakes in the name of authenticity. I want it to understand the difference between something that's part of how I communicate and something that simply gets in the way of what I'm trying to communicate.
That's a much harder problem. It's also a much more interesting one.
Better communication isn't imitation
There's a popular instruction people give AI when the output doesn't feel right: “Make it sound like me.”
I've used it plenty of times. But I've started questioning what we're actually asking for.
If AI perfectly imitates the way I already communicate, it will reproduce things I genuinely want help improving. That's not particularly useful.
At the other extreme, if AI “improves” everything by normalizing my communication into polished, perfectly structured prose, we've solved the grammar problem by creating an identity problem.
What I actually want isn't imitation. It's interpretation.
Understand what I mean. Understand who I'm talking to. Understand my relationship with that person. Understand where I'm communicating. Understand what I'm trying to accomplish.
Then help me express the thought more effectively without replacing the person behind it.
AI loves normalization
This is one of generative AI's superpowers. Give it something unstructured and it can create structure remarkably well. It's also part of the problem.
Messy human thought goes in.
Perfectly structured AI prose comes out.
Do that enough times across enough people and something strange starts happening. Everybody begins sounding vaguely the same.
You've probably seen the voice by now. Perfectly organized paragraphs. Predictable transitions. Thoughtful but suspiciously balanced arguments. Every idea wrapped up neatly before the next one begins.
Nothing is technically wrong with it. But increasingly, nothing is particularly you about it either.
We're using technology capable of extraordinary personalization to create an incredible amount of remarkably similar communication.
I don't think the answer is deliberately making AI writing worse. I don't need random sentence fragments inserted into my LinkedIn posts to prove a human was involved. And if my original thought is confusing, I don't want AI preserving the confusion because that's technically “my voice.”
I want help.
I just don't want the help to replace me.
Humans are more complicated than a tone dropdown
When two people communicate well, we're processing far more than the literal words being exchanged.
Who is speaking? Who are they speaking to? What's their relationship? Where is this conversation happening? Why are they communicating? What are they actually trying to accomplish?
We adjust for those things almost instinctively. “Can we talk?” means something very different coming from your partner at 11 p.m. than from your coworker at 10 a.m.
Same three words. Completely different communication.
Yet a lot of AI-assisted writing still boils communication down to instructions like: Professional. Friendly. Concise. Casual. Human.
Those instructions aren't useless. They're just blunt instruments. Because there is no universal setting for “human.”
Maybe humans shouldn't have to become prompt engineers
This is where the idea gets bigger than writing.
For most of computing history, humans have carried the burden of learning how machines want information. We learned commands. We learned syntax. We learned menus. Now we're learning prompts.
But AI creates the possibility of reversing some of that relationship. The machine can increasingly carry more of the burden of understanding us.
That means the goal shouldn't be teaching everyone the perfect prompt for making AI communicate correctly.
A person should be able to write: “Say this better.”
And the system should be capable of understanding what better means for that person, that recipient and that situation.
Because humans rarely express everything perfectly. I certainly don't.
Sometimes I know exactly what I want to say. Sometimes I know what I mean but can't find the words. Sometimes I start explaining something before I've completely figured out where the thought is going.
That's not necessarily a bug AI needs to remove from me. That's my brain doing what brains do.
The technology helping me should understand the difference between the parts worth preserving and the parts worth improving.
That's the shift I'm interested in
I don't think human communication needs another template. And I don't think the future of AI communication is simply building machines that become increasingly convincing at sounding human.
There's a more useful question:
Not: How would a human write this?
But: How would this human communicate this idea, to this person, in this situation—and how can AI help them do it better?
Because I don't need AI to become more human.
I need it to get better at understanding the human who's already here.
And once you start thinking about communication that way, another popular idea starts looking very different:
“Write like me” might be the wrong goal entirely.