We’re All Prompt Engineers

‍

‍

You may have read my recent article Prompt Engineering Jobs are a Mirage. I argued that there may be a handful of actual engineers working at AI companies to help with the training or QA of LLMs through the use of prompts, but the army of prompt engineers that every company is supposed to be hiring will never happen. I’m now going to argue the opposite is true, at least philosophically.

If you’ve ever raised a young child, you know phrasing is important. “What shirt do you want to wear today?” can lead to five minutes of vacillation whereas, “Do you want to wear the yellow shirt or the blue shirt?” quickly leads to a decision. That’s one of many examples parents know well.

But It's not just children who need this, we all do. This is especially true for managers. Your manager can tell you to do some task any of a dozen ways and you’ll do it. But if you need to make a request of your manager, or you’re a manager who needs to ask for something from someone not in your reporting path, how you ask matters.

“Can I leave early next Friday?” works better as, “Remember how I stayed late last week to finish up those tasks; I’d like to take off early this coming Friday since we’re off to see my parents who live three hours away and I want to head out before rush hour traffic.” This phrasing employs the principle of fairness often used in negotiations. Given what I did for you, I’m asking you to do for me.

I’ve often spoken about how people have communication preferences and styles. If you want to pitch me an idea, it helps to have quantitative evidence and a clear plan. Others are moved by personal stories or visionary ideas more than hard data. (Consider, everyone who understood the real estate market knew that WeWork’s margins just couldn’t work for the returns needed, but the investors bought into the vision.)

How do we get colleagues to buy-in to our ideas? We pitch it a certain way. If I want the CEO to give me the greenlight on a project idea, the best way I can do it is to align it to her initiatives. Instead of asking, “I want to build X; I think it will increase sales,” I know the odds of getting the answer I want go up when I ask, “I want to build X; I think it will increase sales, especially among first time home buyers who I know you set as a new strategic market segment for us to focus on in the coming year.”

This concept is known as choice architecture. It works in subtle ways, like putting impulse purchase items in the check-out lane at stores. It’s Netflix making the next episode opt-out instead of opt-in by having it auto play. We see it with the good-news bad-news pattern or the compliment sandwich technique. In all cases we choose words, formats, framing, and decision paths specifically to elicit the outcome we want.

In my prior article I argued that prompt engineering jobs don't exist. While we will all need to get better at creating good prompts, just like we learned to get better at web searches, it’s a technique applicable to many jobs and not a job in and of itself. The same concept applies here, the of use choice architecture techniques is something many of us find helpful in our jobs. I’d go so far as to say we all do it to some degree even if we’re unfamiliar with the concept formally. We are, to be fair, all prompt engineers; we’ve simply been engineering our prompts for natural intelligence (as opposed to artificial intelligence).

Investing in this skill can help you be more effective at work. As with many skills, it can be employed for good (e.g., working towards good outcomes for your organization or community), or evil (e.g., manipulating people). The skill itself isn’t good or bad, but rather it’s application. Those wanting to learn more can see some of the books I list on my book’s resources page, including Nudge: Improving Decisions About Health, Wealth and Happiness and The Invisible Gorilla: And Other Ways Our Intuitions Deceive Us among others.)