What type of content do you primarily create?
Even if you're already a pro at using AI tools, sometimes figuring out the most effective way of working with them is tough.
Which is why we recently covered a working paper from Harvard Business School that did an AI experiment with 758 consultants from Boston Consulting Group to find the best ways to use AI in the workplace.
You can get all the details in our rundown, but one particular element seemed worthy of its own article. It turned out that the consultants who performed the best—the AI superstars—organized their workflow in one of two ways: they were either a "Centaur" or a "Cyborg."
Here's how you can use these styles too.
Team Centaur
Team Centaur was all about keeping the AI tasks and the human tasks separate—like the body of a centaur, which has clear delineation between the human part and the horse part.
Apparently Midjourney didn't get the memo; it still has no idea how to draw a centaur.
Centaurs would allocate tasks to the AI based on their own human strengths vs. those of the AI. This took knowledge of the capabilities of the tools themselves, which may be why they were so successful. They were much more aware of what tasks the AI tool was good at and which ones it would struggle with.
For example, instead of asking the tool to write a memo itself, a Centaur approach would be to first figure out the main points they believed that the memo should include, and include those main points in the prompt. This process draws on the relative strengths of the human to generate good recommendations based on their experience and analysis skills, and the strengths of the AI to organize and refine the writing.
Centaurs used the AI tools consistently for some tasks:
- Mapping the problem domain: Tasks like asking for background information, explaining terminology, and to generate general knowledge. They then used this information to generate their own analysis and ideas.
- Gathering methods information: They also used it to learn about what methods they could use to complete the task—the "how" rather than the "what." They used that information to either complete the task on their own or to delegate it to the AI.
- Refining human-generated content: Centaurs used AI tools to organize text at the beginning of the process and to make final edits to the text at the end. By doing this, they could improve the writing’s readability, conciseness, and overall quality.
Example Centaur prompt
This prompt shows how a Centaur uses their own background knowledge and analysis to lay out the main points of a memo they’d like the AI to write for them. Importantly, they’re not telling the AI to come up with reasons for driving growth to a particular brand; they’re providing the reasons and just asking the AI to organize them into a persuasive memo.
You are writing a memo to a CEO of a company to inform him of where his company should focus as they attempt to drive revenue growth for one of their three brands, Kleding Man. You want to provide him with the following facts for why he should focus on driving revenue growth for Kleding Man:
- Several of his team members recognize that their brand awareness in this category has slipped and they are no longer serving the older men demographic.
- Their price point is too high for younger men, their current target demographic, which is why they have not been able to even retain constant sales.
- Their share of revenue for Kleding Men has dropped 20% or 11 percentage points, or a 5.44% CAGR.
- Their revenue for Kleding Men has dropped by 22% or a 6% CAGR since 2013, or roughly ~$19M in absolute terms.
- The total % of the clothing market that is made up of menswear is 52%; they are at 44% of their brands indicating there is room to grow and bounce back.
Tips for Centaurs
If you're going to use the Centaur method, there are a couple of things to keep in mind.
Know your tools
Understand the strengths and weaknesses of the AI you're working with. Keep yourself updated with the latest features and limitations, and practice using the tool to see how well it works for a variety of different tasks.
Take your time
When you're using the Centaur model, it can be time consuming to come up with a comprehensive prompt—but that's why it works so well. If you find yourself doing similar or related tasks, starting your own prompt library can help so you don't have to start from scratch each time.
Start small
If you're new to the Centaur method, begin with giving the AI simpler tasks, note where it goes wrong, then redo the whole prompt with the addition of that information. This will help you get a better idea of what you need to include in the prompt in order to generate better prompts, Centaur-style.
Team Cyborg
Meanwhile, some consultants worked much more intensely with the AI tool to refine and co-develop the output. The researchers called these users Cyborgs because they worked as if their minds were merged with that of the AI, using AI on every step of the process rather than doing some parts on their own.
Cyborgs also frequently used advanced prompts like the Persona Prompt in order to get the AI tools to work with them at an expert level. They also tended to question and push back on the AI by asking for elaborations, explanations, and validations.
Cyborgs used AI tools for many more kinds of tasks in their workflow as compared to Centaurs:
- Assigning a persona: Cyborgs often asked the AI to take on a specific persona, such as a marketing expert or business strategist. This helped in aligning the AI’s responses to specific tasks in their workflow.
- Requesting editorial changes: Rather than editing the output directly, cyborgs described the edits and asked the AI tool to make them.
- Teaching through examples: Cyborgs provided guidance to the AI tools in the way of examples so that the tool could emulate the kind of response that the user wanted.
- Modularizing tasks: Instead of asking for one large task, Cyborgs broke the tasks into smaller pieces that the AI could tackle individually.
- Validating outputs: Cyborgs didn’t blindly accept what the AI produced. They would ask the AI to verify its results, adding an extra layer of scrutiny to make sure everything was complete and accurate.
- Demanding explanations: If the AI made a recommendation, Cyborgs asked for the reasoning behind it. This helped to make sure that the tool's suggestions were well-founded.
- Exposing contradictions: When the AI tool made logical or factual errors, Cyborg would point them out to the tool, rather than correcting the error themselves.
- Elaboration: If the AI provided a particularly interesting or unexpected point, Cyborgs would ask for further details. This gave the user the opportunity to decide whether or not it was good enough to include in the final result.
- Directing: Cyborgs guided the AI to focus on particular areas or data points. This allowed them to avoid wasting time on general information and center the analysis on the most important points.
- Adding their own data: If Cyborgs thought the AI was missing something important, they would provide extra data or context. That way, the tool's output lined up with the user's perspective.
- Pushing back: Cyborgs didn't just accept the AI’s output if it didn’t meet their criteria. They challenged it, asking the AI to reconsider its responses. They got a better final product as a result.
That's a lot of stuff—maybe that's why Midjourney thinks cyborgs should be staring intensely off into the distance.
Examples of Cyborg Prompts
Is that all? Can you make sure to revisit notes and make sure you added all important takeaways?
There was a point made in the interview about how men's sales have decreased due to lack of kids' sales. Add that to the above bullets as well.
Tips for Cyborgs
Here's how to make the Cyborg method work for you.
Challenge and question
Cyborgs got the best result by pushing back against the AI’s outputs using their experience and knowledge. This kind of scrutiny can lead to a more robust and nuanced end product. And don't worry—you won't hurt the AI’s feelings by criticizing it (just don't get into the habit of doing that to coworkers).
Know when to restart
Large language models (LLMs) have a limited context window—what us humans refer to as “a memory.” The Cyborg process can lead you down a pretty long prompting chain, which means that the AI tool can forget things that are too early in the conversation. By setting up multiple conversations tailored to each task or summarizing the results periodically, you can get better results.
Always validate
Even if you're asking the tool to explain its logic, make sure that you agree with what it's saying rather than taking the output at face value. Yep, I know that's the hard part. Sorry, not sorry.
Use advanced prompts
Cyborgs often used advanced prompts like the persona pattern and prompts that asked the AI tool to explain its reasoning step-by-step. This allowed them to validate the outputs the tool was providing and access the higher-level abilities of the AI.
So are you Team Centaur or Team Cyborg?
You probably realize that your style tends to be one way or the other. But here's the thing—you can switch it up to see how the experience and results differ. You might be surprised to find you're a bit of both.
Wait—Midjourney, is that a cyborg horse?
