Many assume that writing a good AI prompt is a matter of typing words in the right order, without saying anything too confusing or contradictory, but the philosopher at Anthropic who spends her days thinking about how humans communicate with machines says that thinking is dead wrong. In her mind, the true secret isn’t fancy phrasing or technical jargon. It’s a matter of clarity, intention and realization that you are effectively trying to teach a nonhuman mind to think with you.
She didn’t begin in Silicon Valley. She was from the world of ethics, linguistics and philosophy — realms where each word carries weight and every sentence is meaningful. So when she signed on at Anthropic, she didn’t treat AI prompts as commands or someone else’s thoughts that had to be enacted. She treated them like conversations. And the longer she spent with models, the more she came to see that the best prompts aren’t complicated ones. They’re the honest ones — that’s where people know what they want.
The biggest mistake people make, she says, is that they jump to the conclusion that AI already knows what it is that we want. It doesn’t. It doesn’t read your mind, fill in the gaps or magically interpret what it was you meant when you typed something obscure. Her advice is straightforward: If you’re not sure what the outcome should be, then neither will AI. You’re mumbled her way into getting someone to buy in, and you were either going to get something that was half done or completely off track,” she says.
One of her favorite strategies is to treat the AI as a collaborator rather than an instrument. When she’s posted on something that’s complex, she explains her logic step by step as though she were speaking with a colleague who needs some context before diving in. And she observes the effect immediately. The answers get crisper, more accurate and more in line with what she actually cares about. She says the model listens better when you speak as a human trying to be understood, rather than as a robot attempting to give orders.
She also mentors individuals about the value of specificity. Too-Broad prompts are most likely to fall apart because they’re too broad. “Write me something good” means nothing. “Make a chart to compare these two ideas based on clarity, tone and usefulness” is everything. The model is most effective, she says, when the goalposts are defined — uncertainty is the enemy of good output. The more white space you leave empty, the more guesswork gets thrown in by the A.I.
But perhaps her most surprising advice is to give vulnerability a space in your prompts. If you’re confused, say that. If you’re stuck, admit it. If you’re not sure what to say, let the model know you need some guidance. It is how the models respond to human uncertainty that gives us these devices as in the fact we can't really predict 100% on a machine. It changes the encounter from a demand to a request.
Another one that she often passes on is to separate the task and the style. People have a habit of dumping everything into one chaotic sentence, and the model is left juggling tone, length, purpose and structure all at once. She’d rather pull it apart: tell me first what you want, then how you want it delivered. It’s as if you were ordering at a fancy restaurant — don’t describe how the dish will be plated before you tell ’em what dish you’re asking them to make!
On paper her approach might sound philosophical, but it’s very practical. She thinks that the future of good AI prompting won’t be driven by secret hacks or technical scripts. It will come from people understanding communication at a deeper level — how humans think, how language shapes thought and how precision changes outcomes. Using A.I., to hear her tell it, is simply a new way of practicing old ideas.
Most striking, about her perspective, is how human it is. She’s not fanatical about optimization tricks. She is interested, she says, in the relationship between intention and expression. She believes the best prompts are offered by people who pause long enough to think carefully about what they’re trying to achieve, rather than racing through the task.
And the more she discusses prompting, the more obvious it becomes that if there are going to be great breakthroughs, they won’t come from the machines. They will do so from us — from how we learn to speak clearly, and ask honestly and cooperate with something that thinks different than we do.
At the end of the day, her advice is straightforward: speak to AI as if you were one human talking to another who wants to help you but doesn’t have any context unless you give it. When you treat the model like a human being, not an automaton from the Matrix, everything changes. And then, suddenly, the prompts you write are no longer educated guesses and become understanding.
