How to Use AI in 2025 to Create High-Impact Knowledge Work

In 2025, it will take you longer to read something  and comprehend it than the amount of time it took   to create it. So, if you think finding the signal  amidst all that noise is only going to get worse   with all the AI-generated regurgitation of  posts on LinkedIn and all the SEO-optimized  

blog posts that don't actually contain anything  valuable—well, you're right. We have to move   up the knowledge work value chain, and  luckily, we can do it with AI as well. Let's come back to the fundamentals  of knowledge work and find where we  

can gain leverage. Knowledge work can  be broken down into four steps. First,   we have our input—data, knowledge,  and information that we gather. Then,   we process it to get some sort of output. From the  feedback of the output, we then improve our input.  

Over time, we're going to spiral upwards and  improve the quality of our knowledge work. Now, most people spend the majority of their  time thinking about how they can outsource   their thinking process to AI. They are  using tools like ChatGPT and saying,  

"Okay, you write this for me;  you do that for me." However,   we realize the results are not particularly  fantastic. Why is that? It's because the   input that large language models (LLMs) have  is pretty average. We're averaging down to  

the level of the Internet; there's some gold  in it, but the majority is pretty mediocre. Instead of focusing on the process, we  want to focus on using AI to improve   our inputs. Even if nothing changes  in the way we process information,  

our outputs will be better because we won't  have this "garbage in, garbage out" problem.   Let's concentrate our efforts on using AI  to help us improve our inputs. The addition   we are going to make here is to incorporate  academic papers and research into our input.

Now, you might think, "First of all,  that sounds boring; second of all,   this is practical work." This is not some sort  of academic research that we're conducting;   we need practical outcomes. But think about  it: Do you read the Harvard Business Review,  

McKinsey Quarterly, or the MIT Sloan Review?  All of these aim to take boring, hard, complex   academic papers and make them  relevant to working professionals,   delivering insights in a way that allows us  to extract something actionable from them.

We need to realize that we don’t want to  be readers of HBR. We read so slowly. If   you just feed the entire archive to ChatGPT,  it will immediately know so much more than   you. We don’t even want to be the writers for HBR  because people spend their entire lives writing a  

few insights. Instead, you can leverage those  insights. Why spend the time researching? What   you want to be is HBR itself for your team or  organization. This means going through trusted,   complex, and nuanced resources, translating  them into the context you’re in, creating  

frameworks that turn theoretical knowledge  and insights into something actionable,   and delivering it in a way that  busy people can understand. We need to move up the value chain and look at  things that used to be reserved for academics,  

scientists, and authors who had the  resources, time, and brainpower to   sift through long-winded papers. Now you  can too. Let’s take an example and let me   show you the three tools that I use to ensure  I become the HBR for my team and organization.

Let’s use an example of one of the  biggest challenges that managers face:   the problem of a team underperforming. There  are bright people, but they seem like they don’t   want to do anything; they’re always complaining  about stuff. How do I improve team performance?

Usually, if you get average input, you might go  to ChatGPT and ask it this question, which will   give you some average results. You might go  to Google, and it might give you some average   inputs. Let’s give it a try. “How to improve  team performance at work?” Common question,  

right? Oh gosh, it tells you to focus  on goals and open communication. These   are just such empty, broad recommendations  that are not going to lead to great output. What I like to do instead is to use the first  tool of today: Elicit. This tool helps you find  

academic papers related to the question you ask.  So let’s copy and paste this and see what kind of   results we get in Elicit. It's free; you just have  to sign up, and you’ll get something like this:   a summary of the top four papers for free,  along with a list of relevant papers.

Now, I like to ensure that what I’m  reading is well-cited and peer-reviewed,   not just some random paper.  I sort by the most cited.   For example, "team design features and team  performance" seems to show a relationship.  

Let’s take a look at this one. What you  can do is click into Semantic Scholar,   which often leads you to the actual paper itself.  In this case, it is locked, so if you don’t have   access to a university subscription or something,  you might not be able to see it. No problem;  

we’ll just copy and paste the paper name and go  into Google to see if we can find a PDF version. Here we go: here’s the PDF published by  Sage. Let’s double-check—it seems to be   the right one. Okay, we’ll download this, and  then we’ll use the second tool of the day:  

Notebook LM by Google. Unlike ChatGPT or Claude,  this tool isn’t using AI on knowledge from the   internet; it only uses sources that you  upload. So, you have a lower likelihood   of hallucination. Sign up with your Google  account, and you’ll see something like this.  

You want to create a notebook, which  essentially just houses your sources. Since we’ve downloaded a PDF, we’re going  to upload it here. Now, we haven’t even   read the paper; we don’t really know what’s  in there. So, what I like to do is use it  

as a table of contents first, which is  automatically generated by Notebook LM. Now we have a table of contents.  "Groups" is important; there’s   classification theory. It looks like  we have to look at characteristics,  

some team size, team level, task design,  and leadership. Okay, interesting. Now,   I don’t want to read this yet; it’s a lot of  text and really poorly formatted. But you know,   Notebook LM is still pretty experimental, so  we’ll give them some time to figure this out.

Now that we have a general idea, let’s  click back on the Notebook guide and   look at the suggested questions. Since we  haven’t read it, but AI has processed it,   I usually like to use one of the questions  just to get a basic gist. Here we go:  

“How do team design features impact overall  team performance?” Let’s see what it says. Looks like there are about three components, and  we saw that in the table of contents as well:   team composition, task design, and leadership.  I love this about Notebook LM. You can click  

into this to find the actual place in the  source where it discusses this idea. So,   even though this giant paper is probably tens of  thousands of words, I can identify which parts   are most relevant to me and actually find the  citation so I can better understand what they say.

Let’s say, for example, that I can’t  change the team composition—I’m not   going to fire anyone or hire  anyone new—but I want to focus   on task design and leadership. I can  prioritize these cited areas first.

That’s the first cool part about the Elicit to  Notebook LM workflow, but this is only one paper.   What I would do again is go back to  Elicit and now see if there are any   other interesting papers. “Leaders can learn  about teamwork by developing high-performance  

teams.” That seems interesting. Let’s take  a look and see if we can find the paper. Oh, here’s the PDF—fantastic! Now  we don’t have to... oh no, okay,   bad request. That’s all right; it happens.  Let’s go back and check the publisher. Oh,  

it’s a Wiley paper. Let’s  see if we can get the PDF. All right, we can’t. If you’re at work, surely  your workplace will help subsidize getting this   paper. But even if they don’t, let’s see if we  can find it somewhere else. Let’s see if this is  

the right one. Fantastic! Here’s the second  PDF we got. We’ll head back to Notebook LM,   and this time we’re going to press  the plus sign and again upload it. Once that’s done processing, you’ll  see down here that now you can chat  

with both sources. But before we do that, I  want to uncheck the previous one because I   just want to understand what this new  paper is about. So let me generate a   table of contents for this one source, and  we’ll see that it’s generating a new note.

Let’s wait for that. All right, here  we go: there are five essentials to   building high-performance teams:  capable and committed members,   norms, a team structure that has clear mission,  responsibility, goals, and a framework to get  

results. See, this is already so much clearer  and more specific compared to the Google   search results, which just got diluted  over time by everyone—maybe one person   starting from the source, but eventually  secondary sources dilute the message.

Okay, a way to manage team processes,  and that’s the fifth one. Great! Now   let’s select both sources and ask, “What are  the key changes a leader can make in a team   to improve performance?” Now it will look  through both sources and give us an answer.

All right, it says: “Focus on member selection.”  Let’s say that’s not relevant to us. “Promote task   meaningfulness”—this was mentioned in the  first source. Great! “Empower the team with   autonomy.” I don’t know if we’ve seen this.  Okay, this was in the first document. Great!

It gives us the exact things we should be  doing: “Cultivate team coordination.” All right,   “Provide effective leadership,”  which means guiding, motivating,   and supporting the team. Let’s see the details.

Now, let’s say out of all of this, I think for  my team the most sensible action is to promote   task meaningfulness. I want to empower the team  and facilitate team development, and of course,   provide effective leadership. So, give  me a detailed plan with three changes  

we can implement to improve team performance. Now you have a high-level plan for doing this,  but it still feels somewhat vague. Even though   we know the input is good and we’re focusing on  what academics have found out, we want to do the  

next part. We’re going to copy this and go into  Claude and say, “Help me create a detailed plan   to improve team performance using…” and I’m going  to paste in what Notebook LM has shared with me. In just two seconds, we got a 12-month,  three-phase, four-module plan to improve  

team performance. Just think about it—if  you were in HR or you were a team lead and   you shared this plan with management, they  would be blown away. But that’s not all we   need. Let’s say, out of all of this,  it still sounds somewhat complicated.

So let’s ask, “What are the 80/20 changes  I can make to my team to drive performance   that depend on leaders and not on the team  members to make a change?” I’ll ask it to   refer to these two studies, which are called  a “meta-analytic review” by Greg and by War.

Now Claude has a better idea of  where I’m getting these insights   from. Based on the details it knows  about these two sources, it gives me   the core leadership action plan. Okay, “Task  design optimization” seems to be one of them,  

“Authority distribution” is the second  highest impact, and “Contextual support.” So now we’ve gone from a 12-month plan  to a 3-month plan, and it gives us some   expected results. Whether that’s true or  not, we can wait and see. But it helps  

us list out things like key success  factors and measurement frameworks. Let’s say, “Help me create a matrix to  better design tasks with clear goals   and interdependence structures.” Here we  go! Now we’ve got a framework that we can  

use to assess goal clarity levels.  We can use interdependence levels,   see where they intersect, and even show the  team, “Hey, this is what I’m thinking about   for changing how we think about tasks. What  do you think?” and get their feedback on it.

Now, if we simply ask Claude, “Help me with what  I need to improve my team’s performance,” let’s   see what we get. All right, it’s asking me a few  questions. That’s actually already an improvement   since previous versions of GPT. Let’s say,  “Work with teams that are underperforming  

and unmotivated.” All right, and see the  plan it gives me is so much more vague. Right? It’s “Communicate and provide feedback.”  I know this! “Having clear goals.” I know this;   I know all of this. But it doesn’t help.

Help me create a framework to set goals.  Yes, SMART is good, and monitoring is good,   but as you can see, because  the input quality is not there,   we’re stuck with these buzzwords  that we hear all the time.

I am excited and optimistic about the changes  coming to knowledge work. We can critically   think more; we can actually be more creative  and associative. Generalists will have a great   time in this era of AI. Let me know how you  feel about it. Check out this video on how I  

use AI to create McKinsey-style visuals,  and I’ll see you in the next video. Bye!