#593 AI Co-Thinking Part 1: Why Context Comes First | Nathan Mauck, Robert Greene, Niels Brabandt
AI Co-Thinking Part 1: Why Context Comes First
Back by popular demand: after our first conversation about AI co-thinking received a remarkable amount of feedback, both authors are now with us for a four-part series. My guests are Prof. Nathan Mauck of the Henry W. Bloch School of Management at the University of Missouri-Kansas City and Robert A. Greene, founder of OIC Innovation, co-authors of AI Co-Thinking: A Framework for Working with AI. In this first part, we look at the most misunderstood word in the current AI debate: context.
A word that means something different to everyone
Context sounds simple until you ask people what they mean by it. Nathan Mauck recently attended an AI conference with six major vendors. All of them talked about context engineering, and all six meant something different. For some, it meant putting all company data online. For others, it meant describing internal processes.
Robert Greene and Nathan Mauck define it methodologically. Their AI co-thinking framework starts with a phase they call customising context, which they see as the foundation for thinking with AI rather than treating it as a vending machine: prompt in, answer out. Context in their sense depends on the problem you are trying to solve. It includes internal organisational information, external data, your way of attacking the problem and your definition of what a successful outcome looks like.
The problem is part of the context
What could be more obvious than defining the problem? Yet Robert Greene has seen individuals and companies cut problem definition short again and again. The intention is often good: nobody wants analysis paralysis, so people want to get moving. The result, however, is that teams start before they understand the connection between what they do and the outcome they want.
One of the most valuable side effects of AI co-thinking, the authors argue, is that thinking with AI forces you to understand the why behind the what and the how. You end up with a better-defined problem, and instead of repaving the existing path, you may find a new and more effective route to the outcome.
How much context is enough?
The answer is a Goldilocks principle: not too much, not too little. If you upload a huge document, the AI does not know how to prioritise the information and returns generic output. If you give it too little, you also get generic output. The work lies in identifying the key ingredients as precisely as possible without overloading the system with detail.
Robert Greene puts it bluntly: too much information is a problem, and it is on you, not the AI, to sort out what is relevant. His practical advice is to err on the side of too little, ask the AI to tell you what else it needs, and add bricks to the informational foundation step by step.
A conversation, not a vending machine
Expect more than one attempt. AI co-thinking is a conversation. Defining the problem is iterative, finding the right internal information is iterative, and so is the most underestimated step: making hidden assumptions explicit. Separate what you know as fact, what you assume and what you do not know. All of this is part of creating context.
What if you know nothing about the topic?
The authors use a two-by-two grid of expertise and stakes. When your expertise is low, AI co-thinking is still useful, but not for jumping straight to an answer. Use it to frame the problem, build expertise, find the right questions and identify which colleagues you need to involve. Narrow the problem, keep your own input modest, ask the AI to ask you questions and give more weight to external information. The one thing they do not recommend is doing nothing because you know nothing. Understand the risk of your lack of knowledge and calibrate accordingly.
Why this matters for decision-makers
For executives, the lesson goes beyond prompting technique. The quality of AI output reflects the quality of thinking that goes into it. Organisations that invest in clear problem definitions, explicit assumptions and curated information will get significantly better results from the same tools than organisations that simply roll out licences. Context is not an IT task. It is a leadership task, because it requires decisions about what matters, what success looks like and which information can be trusted. Leaders who model this discipline in their own use of AI set the standard for their teams far more effectively than any training programme or usage target could.
Three takeaways for leaders
1. Define the problem before you define the prompt. Context starts with the problem and the outcome you want.
2. Curate, do not dump. Relevant context beats maximum context.
3. Make assumptions explicit. Separate facts, assumptions and unknowns before you let AI build on them.
This article is part of a four-part series with Nathan Mauck and Robert Greene on AI co-thinking. In Part 2, we look at why every specialist should become a better general manager. I wish you every success in applying this in your organisation.
Niels Brabandt
---
More on this topic in this week's videocast and podcast with Niels Brabandt: Videocast / Apple Podcasts / Spotify
For the videocast’s and podcast’s transcript, read below this article.
Is excellent leadership important to you?
Let's have a chat: NB@NB-Networks.com
Contact: Niels Brabandt on LinkedIn - https://www.linkedin.com/in/nielsbrabandt
Website: www.NB-Networks.biz
Podcast and Videocast Transcript
Niels Brabandt EMBA MBA MSc: AI co-thinking and back by popular demand. Now both authors are with us. We received a lot of feedback from you and you loved the episode. So here they are again. We're going to talk about AI co-thinking and much more than that. Hello and welcome Robert Greene and Nathan Mauck.
Robert Greene: Niels, pleasure. Good to be with you.
Niels Brabandt EMBA MBA MSc: Brilliant. Thank you very much. So in your book, you talk about the importance of context. The main problem is with context, many people find this word very ambiguous. They think, is it my idea is it is it my approach is it someone else is it the Organizational approach so, When this is an ambiguous term how in your opinion can this in any way be related To a real world project so can you give us a bit more on on this context term.
Robert Greene: Nathan, let me jump in and maybe build on the start that you made in the last episode with Niels. And context to us is defined methodologically, in fact. We have an AI co-thinking framework that we propose, a simple framework, A commonsensical framework. But its initial phase is all around establishing context, which is, We believe, the foundation to thinking with AI as opposed to treating it as A vending machine and just using AI. And it has three components in this phase one, which we call customizing context. And I don't want to monopolize here, Nathan, so let me shift over to you.
Nathan Mauck: No, it's great background. And it's interesting, you mentioned this context being confusing. I was just at an AI conference with some of the big vendors in AI, And context engineering was the phrase. But it meant something different to all six vendors. You know, For some people, it meant let's get all the data in the organization online. Let's let's talk about our process for solving problems. And, you know, I think Bob and I really come at it from a place of the context Depends on the problem you're trying to solve. And so it's all the context that will help you get the best solution, Whether that be internal information. So it could be internal organizational data, external data could be your problem Solving methodology, your way of thinking about attacking the problem, Your definition of what a successful outcome looks like. And so I think. It's sort of driven by the problem is the key, and it kind of cuts through some of that confusion.
Robert Greene: Yeah, and just building on Niels, it's interesting. Well, context includes the problem. Well, what could be more obvious? But I've been amazed over the years to see how often individuals and companies Kind of short shrift problem definition. They kind of they kind of and it's in large part, it's often a good sign. They're just so analysis paralysis is bad. Let's get moving. Let's get started. Let's figure this out. Well, Oftentimes you get moving before you truly understand what is the connection Between what you're doing and the outcome that's desired. And in fact, a side benefit, if not a primary benefit of AI co-thinking that We find is often through the process of thinking with AI, one better understands The why driving the what and the how. So ultimately, you get to a better outcome, but also you start eventually with A much better defined problem and a better, which drives, you're not repaving the existing path. In fact, you're maybe finding a new route to that outcome that's more efficient and effective.
Niels Brabandt EMBA MBA MSc: Yeah, excellent. And when we talk about context now, of course, You can say, hey, this is part of the context. And I think we should include this one as well. And the more people you ask, The more context you will get. And you will probably end up questioning your whole business model at the end Of the day. So the main question now is, how do I know how much context is enough For best or can we say optimal? How can we define what is the right amount of context with not going too little or too much.
Nathan Mauck: There's a little bit of a Goldilocks, not too hot, not too cold, Right in the middle to all of this, and that you want the relevant context. But one thing about, let's say, how AI works is if you give it a ton of context, You download a huge document, it doesn't really know how to prioritize all of the information. And so you run the risk of if you overload that you just get kind of generalized generic output. But also you run the risk if you don't give it the right amount of context enough Context you get generalized kind of generic output yeah so so you know you really Have to to get to the the key ingredients um Uh is is kind of deeply and essentially as you as you can um and and then not Overload with um with excessive detail
Niels Brabandt EMBA MBA MSc: Would you say that we should expect to need more than one try to get to the best result?
Robert Greene: Well, yes. And again, Nathan and I would like to use this counter metaphor as A vending machine, prompt in, answer out, but let's move on. AI co-thinking is very much a conversation. And as I suggested a moment ago, Even figuring out what the problem is, is an iterative process. And figuring out what the proper context, what the proper internal information Is, can be an iterative process. Understanding what your assumptions Are, extracting the hidden assumptions and making them explicit. Understanding what the facts are, what your assumptions are, and what you don't know. All of this is part of creating context and just reinforcing what Nathan said a moment ago. TMI, too much information, is a problem. Don't expect AI to sort through your informational debris. It's on you to figure out what's relevant. Probably makes sense to err on the Side of too little while asking AI to ask you what is relevant here and start Adding bricks to the informational foundation.
Niels Brabandt EMBA MBA MSc: Excellent. And, of course, one last question for this episode I need to have. Let's say people face a question where they say we never had this. So people expect me to prime my AI in some way. And I said, I have no idea. We never had this. So here's the problem. And I know exactly as much as you now. Could you please help me how to do that with the right context when you by yourself Say I have no idea where to start?
Nathan Mauck: We think AI co-thinking works in kind of all scenarios, but we do like to think About this two by two grid of expertise and stakes. So high expertise, Low expertise, high stakes, low stakes. And that low expertise one, we still think AI co-thinking can be really helpful, But it's sort of not ideal to ask, just jump in and ask for an output or result or an answer. It's more of help framing the problem, maybe help building expertise, Help figuring out what are the right questions to be pondering over, And really even kind of identifying maybe which other humans do you need to Bring in to help you get the right expertise.
Robert Greene: I Think these are the perfect final words. Yes, Robert. I think one can even think Of calibrating relative to your good question around the three components associated with context. Maybe if you're not sure, maybe you narrow your problem. If you're not sure, you kind of minimize your input and internal information And you request that the machine ask you and you backfill with folks outside In other parts of the organization. That can amplify, or you weight more the external information. What we don't propose is if you know nothing, don't do anything. Just understand that that's the risk associated with your lack of knowledge And calibrate and adjust the contexting accordingly.
Niels Brabandt EMBA MBA MSc: I think these are the perfect final words. So you see, calibration is key. The question is now, why do some people say everyone should be a better general manager? And we're going to talk about exactly that in our next episode so for today At the end of this episode there's only one thing left for me to say Nathan And Robert thank you very much for your time thank you.