Last issue I wrote about crowd wisdom, in Are Right, A Lot: crowd wisdom with humans and AI, and described large language models as a compressed version of that crowd. Here I turn to your side of it: what you bring to the crowd, and how that helps you get much more out of it.
Why do two people get different results from the same AI?
Hand the same task to two people using the same AI tool, and the results can look very different. We tend to pour our deep beliefs and expertise into our prompts, along with our biases, mostly without noticing. I saw this happen in my recent vibe coding workshops, where the app each participant ended up with reflected a lot of what they knew and cared about.
One way to explain that gap comes from Ethan Mollick, the Wharton professor behind the One Useful Thing newsletter. In his new book, Co-Existence, he names four human advantages for the age of AI: deep knowledge, wide knowledge, taste and agency. His image for it is a vast library. The AI holds all of it, yet many people only manage to find the books they already know are there.
Each of these potential advantages lines up well with one of Amazon's Leadership Principles, and each pairing comes with something practical you can try.
How does deep knowledge connect to Dive Deep?
Deep knowledge is the kind of expertise that lets you spot quickly that something is off. It helps with AI in two ways. You notice where the AI gets your field wrong, and you ask better questions. A novice asks something generic and gets a generic answer back, while an expert adds the constraints and edge cases that make the result stand out.
The matching principle is Dive Deep, where leaders "stay connected to the details, audit frequently, and are skeptical when metrics and anecdote differ." Expertise is what allows you to audit the AI's answer in an area you know well, and it works even better when you give it to the AI before it starts:
Here is my task: [task]. Before you answer, list the constraints, edge cases and assumptions an expert in [field] would check, and ask me about any you can't infer.
Mollick also notes that AI now does much of the junior work through which experts used to learn their field. For anyone leading a team, that is a Hire and Develop the Best question: where will your next experts come from? Hold back some of that learning work, along with your feedback on it, for the people still building their expertise.
How does wide knowledge connect to Learn and Be Curious?
Mollick explains that AI has learned from a huge range of fields. His examples run from design thinking and Bayesian reasoning to the Toyota Production System and Rogerian therapy. But it may not offer you a method from another field unless you ask, and you can only ask for methods you know exist.
This is Learn and Be Curious, where leaders "are curious about new possibilities and act to explore them." There are two ways to widen what you ask for: learn outside your own field, and ask the AI to bring in methods from fields you don't know yet:
Which frameworks or methods from other fields could help with this problem? Name three, explain each in two sentences, and tell me which one you would try first and why.
How does taste connect to Insist on the Highest Standards?
Producing a draft or a product concept took real effort before AI, and so did writing a paper. Now building and making is quick and cheap, and in Mollick's words, "The scarce resource is no longer the ability to produce." The scarce skill is looking at a pile of output and knowing which piece is good. Back in issue #35 I noted that a system can run a hundred AI jobs at the same time, but one person can properly review the outputs of perhaps three or four. Taste is how you tell which of those hundreds of artifacts are good, and decide what to do with them.
The matching principle is Insist on the Highest Standards. Jeff Bezos wrote in his 2017 letter to shareholders that high standards can be taught, and that in any domain you first need to "recognize what good looks like." So taste is something you can build together with your team, and you can practice it with AI or with your team:
Give me ten different versions of [the output]. I will pick the best one and tell you why. Then use my reasons to write ten more.
When you practice it with your team, record the reasons behind each choice. Over time they can add up to your team's own description of what good looks like, and you could even turn them into Team Tenets, the topic of issue #38.
How does agency connect to Bias for Action?
Mollick says agency may be the most important advantage of the four, and also the hardest to talk about, since it is difficult to define: the willingness to push and experiment, and to act without waiting to be told. With AI it takes on a specific meaning. It means testing what the tools can do when nobody can tell you ahead of time where the limits are, and those limits keep shifting as AI improves and can do even more.
Here the principle is Bias for Action:
Speed matters in business. Many decisions and actions are reversible and do not need extensive study. We value calculated risk taking.
Many AI experiments are the reversible kind of action this principle has in mind, with no need for extensive study. Even so, Mollick describes research where nonexperts assumed that when the AI failed at something on the first try, it simply couldn't do it. In this context, agency means trying again, in a different way. Some ways to build it in yourself and in your team:
- Try twice more before giving up. When the AI gets it wrong, add context, split the task into steps, or ask what it needs from you: "I tried to [task] and the result wasn't good enough because [what went wrong]. What information or steps do you need from me to do this well? Ask me before you try again."
- Run one experiment nobody asked for. Choose a task from your own job that you haven't tried with AI yet, and try it.
- Make quiet experiments visible. Many people in your organization may already be using AI on their own, without telling anyone. Give them fifteen minutes in a team meeting to show one thing that worked, so one person's agency turns into something the whole team learns from.
- Show your own experiments. As a leader, tell your team what you tried with AI, including what didn't work. That shows them experimenting is part of the job.
Go deeper
For a deeper dive, watch Ethan Mollick in conversation with Brian Keating on the "Into the Impossible" podcast, about what AI can do.
Your action step
Pick the advantage you would like to build next, and use its prompt from this post on one real task this week. If you pick agency, start with the experiment nobody asked for, and show the result to your team.
Which of the four advantages is your team strongest in today, and which one do you want to build next?
If you are working out how to build these advantages in your teams, that is the kind of question I take on in AI strategy advisory engagements and in sessions as an AI keynote speaker and workshop facilitator.