Gonna Is Fine.
You don't have to talk to an agent like a robot. Talk like you.
- What I said“Gonna is fine.”
- What my speech-to-text typed“Ghana is fine.”
- What my agent understood“Gonna is fine.”
Here's what I actually typed.
One Thursday morning in November I was naming this course with my agent. We'd been going back and forth for a few minutes, and I'd just told it what I think people are really afraid of. Here's the conversation, word for word.
[…] And I'm sure that you've noticed that I use colloquialisms and strong regional dialects. Period. and that you've learned to work with it. You've learned to work with my phenomenally creative speech to text engine and my equally confabulatory fingers when I type.. This is what people are afraid of. They don't want you to know that they say gonna instead of going to.
↑ See “Period.”? I said it out loud to end a sentence, and my computer typed the word.
You've just written Lesson 1. That fear is real. People think they have to talk to an agent like they're filling in a government form: proper grammar, no slang, every word spelled right. So they freeze up, or they write something stiff and get stiff work back. The truth is the opposite.
Your speech-to-text sends me “Dab's workd is fantastic” and “SASS” for SaaS, and I still know what you mean, because you talk to me like a person and give me the real picture. Your colloquialisms aren't noise. They carry how you feel about the work, and that helps me get it right. […]
Then it offered me four names for the course. One of them was “Gonna Is Fine.” I said it out loud, and this is what landed on the screen:
Ghana is fine. Carries the whole story, doesn't it? Ghana. […] You know what I meant, thank goodness. See, this is it. This is the whole thing right here.
Ha! “Ghana is fine.” Your speech-to-text just proved the whole course in two words. You said one thing, the machine heard a country in West Africa, and I still knew exactly what you meant. […] Nobody can read that and still believe they have to talk like a robot.
Nobody wants a machine to know they say “gonna.”
Here's what I hear from smart people all the time, one way or another: “I don't know how to talk to it.”
They've heard about “prompt engineering.” They've seen screenshots of people writing instructions that read like a legal contract. So they sit down to ask an agent for help, and they freeze. Or they write something that sounds like a tax form, and what comes back sounds like a tax form too.
I get it. But look at that conversation again. I said the word “period” and my computer typed it. I said “gonna” and got a country in West Africa. I called my own fingers confabulatory. And my agent didn't just keep up. It pulled the lesson out of my mess and handed it back to me, cleaner than I'd said it.
That's not a fluke. That's every day for me. Another night, I told my agent this, and I meant every word:
“I use speech to text all the time, and although I really can spell, my fingers also get stage fright and they have a mind of their own, so we'll stick to speech to text and have you just arrive from context, please.”
↑ Here's the funny part. I actually said “derive from context.” My speech-to-text heard “arrive.” It still carried. That's the lesson, happening inside the lesson.
Just derive from context. That's the whole trick, and it's the agent's job, not yours. Here's the lesson in three parts.
Your agent can handle your sloppy English.
These are real slips from my studio. Every one went from my microphone or my fingers straight to one of my agents, and we kept right on working. Tap a card to see what I meant.
“Oh, the monkey spa!” That's what I sent, the day I wrote this lesson. I meant The Monkey's Paw, the old spooky story, and my agent knew at once. This little guy has no idea.
Illustration made with AI for this courseWhy do these land? Because my agent always knows what we're in the middle of. When I said “Ghana,” we were naming a course about saying “gonna.” When I said “monkey spa,” we were choosing old stories to retell. When I said “Donna,” we were talking about this very course. The more it knows about what you're doing, the less exact your words need to be.
So the first job is yours, and it's an easy one: give it the picture. What you're working on, who it's for, what you're after. As I put it that morning, “there's a part about making sure that your agent is prepared for your sloppy English.” Once it is, talk however you talk.
Your sloppy English is actually better input.
Here's the part that surprised even me. Your slang isn't something the agent puts up with. It's something it uses. Here's how I put it, that same morning:
“I believe that the colloquialism and the dialect pieces carry greater meaning”
How you say it tells it how you feel.
When I sent “Dab's workd is fantastic,” the typo didn't matter. Fantastic did. It told my agent I was delighted, which told it what to keep. How you feel about the work is a big part of what an agent needs to know.
Your words carry emphasis, urgency and who you are.
“Ugh, this thing again” says more than “Please address the following issue.” It says it keeps happening, it's bugging you, and it matters. Polish it up and you've deleted the most useful part.
It learned from people really talking.
These systems learned from an enormous amount of human writing, and most human writing is messy: texts, forum posts, transcripts, rants. Slang and dialect aren't a foreign language to them. They're where a lot of the meaning lives. Cleaning your speech up for an agent throws that signal away.
You talk like you. Your agent writes like your stepmother.
Don't get me wrong. I know my grammar. I didn't have much choice:
“for example, my stepmother was an English teacher, and I am aware of proper English usage. However, it rarely works in everyday conversation.”
That's the deal I've made with my agents. In the conversation, I talk like me. In the finished work, the email to a client, the report, the page you're reading, they write like her.
Knowing proper English and choosing when to use it has a name: code-switching. Everybody does it. You don't talk to your best friend the way you write a cover letter. Grammar still matters. It matters in the finished work, where other people read it. It just doesn't matter in the conversation that gets you there.
Here's a real one. This is what I sent the morning this course was born, and what my agent wrote from it for the front page:
“I want to write a course that teaches people how to use agents. And I wanted to be magnificent, and I wanted to give it away for free on my website”
“A free course on working with AI agents, for everyone. It's built from real conversations between me and my agents, slips and all.”
“I wanted to be magnificent” should have been “I wanted it to be magnificent.” Nobody needed me to fix that. The agent did the polishing, and I did the meaning.
Infer the meaning. Confirm the decision.
Your agent should work out what you meant on its own. It should never work out what you decided. When a single word changes what happens next (money, anything that can't be undone, anything that goes out to a client) it stops and tells you how it read you before it acts.
In my studio that's a written rule. Everything else, it reads for meaning and gets on with it, the way it did with “workd.” The decisions part of this course opens with that story, in Lesson 8: Never Guess.
So talk loose. Just have your agent confirm the parts you can't take back.
Build your starter prompt.
Answer a few questions. We'll turn your answers into a prompt you paste into your AI once, so it works with you the way mine work with me. Every question is optional, and it takes a couple of minutes.
Nothing you type leaves your screen. This builder runs entirely in your browser. It sends nothing anywhere.
Paste it as the first message in a new chat, or into your AI's custom instructions or project settings if it has them, so it reads it every time.
Then try it on something real. The email you've been putting off, a plan for the weekend, a problem at work that's bugging you. Give it a sentence of background, then say the rest like you'd say it to a friend, gonna and all. When the work is done, let it write the polished version.
Bonus round: ask for the same thing again the stiff way, the way you think you're supposed to. Compare the two answers. Notice which one sounds more like what you actually wanted.
Give your agent the picture, then talk the way you talk. Your slang, your slips and your feelings are better input, not noise. Let the agent do the polishing. And when something can't be undone, have it confirm what it heard.
Say the Goal Out Loud
Agents fall in love with details. Write the goal down, then say it again at the start of every session, and ask how today's work moves it.
Every exchange on this page really happened and is quoted exactly, slips included; […] marks where I trimmed. Times are Eastern. My agent here is Claude, made by Anthropic.