You may have heard someone in tech say that “prompting is dead.” Usually the argument goes something like this: stop telling AI how to do the work. Give it the goal, press enter, and let it figure out the rest. And honestly, for some kinds of work, that may be right.
If you have a clearly defined technical task and can describe the result you want, today’s models can often do a surprising amount on their own. That is useful. It is also a fairly narrow view of what people actually do at work.
A lot of professional work does not start with a goal and end with a finished product. Learning is a good example. For fun, open a chat window and type: “My goal is to learn everything about astrophysics.” Press enter. You will undoubtedly get something about astrophysics. You will not, however, have learned everything about astrophysics. I am reasonably confident Neil deGrasse Tyson’s job is safe for another day. AI didn’t fail in that example. That’s just not how learning works.
So let’s talk about a few kinds of work where prompting AI is still important and, I’d argue, critical. To be clear, I’m talking about an iterative process. That just means having a conversation with AI about a topic over multiple exchanges.
Learning is iterative
When you are trying to understand something complicated, an answer usually creates another question. Sometimes you need an example or the same idea explained differently. Sometimes you discover that the thing confusing you is three layers underneath the thing you originally asked about.
AI can predict some of that. It can build a curriculum and anticipate some questions. What it can’t know is what you don’t understand until you start interacting with it. That interaction is prompting.
Writing is iterative
Writing is another example. Ask an AI for a two-page report on the meaning of life and you will absolutely get one. It will have paragraphs. There is a decent chance it will contain words like “real,”, “actually”, “complex,” and “quietly.” It will have flawless grammar. That doesn’t mean it is good. Chances are it won’t be. No offense, Claude.
Good writing involves reading words and deciding whether they work together. That’s judgment. AI doesn’t exercise judgment the way a person does. It generates language from patterns, which is one reason its writing can become repetitive, bland, or miss the point. It also loves to hear itself talk, so the drivel is LONG. Again, no offense intended Gemini.
Your junior high English teacher was right: you need an outline before you write anything good. Why? Because otherwise your writing can become impossible to follow. An outline helps guide the process but you’ve really got to read what you wrote and decide: What are you really trying to say? What should be cut? Where did the argument wander off? Is another paragraph sitting there because you didn’t have the heart to kill it? You usually cannot answer all of those questions before the first draft exists. And guess what? Neither can AI. It’ll write something, but how can it answer those questions? YOU are the one (allegedly) who’s writing something. If you don’t know, how can AI know?
I write with AI all the time, and some of the most useful prompts I give it are things like:
“That paragraph doesn’t belong.” “No. Try again.” “That isn’t my argument.” “You used ‘real’ and ‘actually’ five times each in a paragraph!”
None of these are especially impressive prompts. Nobody is going to buy a $499 masterclass built around “No. Try again.” They are useful because I am reacting to the work as it develops. That’s prompting.
Strategy is iterative
Strategy is one of those words corporate America has managed to make sound far more mystical than it is. People talk about it as though a Strategy is hidden somewhere in the building, glowing softly in a locked room, available only to senior executives and McKinsey alumni.
At its simplest, strategy is figuring out how you are going to achieve something over time.
The hard part is that time passes. Things change. Competitors react. People behave differently than expected. Budgets move. Assumptions turn out to be wrong. Somebody points out a constraint nobody considered. One new piece of information can change several other parts of the plan. So you revisit it.
Those changing dynamics are where AI can be enormously helpful. You can ask it to challenge an assumption, look at the plan from someone else’s perspective, find dependencies you missed, argue against the idea, or work through what happens if one part of the plan changes.
Then you look at what it gives you and decide whether it makes sense. Sometimes it will. Sometimes it will completely misunderstand your business and you’ll have to correct it.
That is not a failure of the process. That is the process. That’s prompting.
What kind of prompting is dead (or maybe was never alive)
Where I think the “prompting is dead” crowd has a point is with the absurdly elaborate prompts people were circulating a few years ago. You probably do not need to spend 700 words telling ChatGPT that it is a world-renowned expert with 25 years of experience, six graduate degrees, and the accumulated wisdom of civilization before asking it to review a spreadsheet.
Models are much better at understanding intent now. Agents are getting better at pursuing goals. We should have to explain fewer mechanical steps. Good. But that is very different from saying the conversation is no longer useful.
For simple, well-defined work, giving AI the destination may increasingly be enough.
For learning, writing, strategy, and other work where the answer emerges as you think, the back-and-forth is part of how you get to something good. So when someone tells you prompting is dead, I would not dismiss them. They may be completely right about the kind of work they do. I would just be careful about assuming they are talking about yours.
We teach two courses on prompting. Prompting 1 and Prompting 2.