Systems Thinking with GenAI: Solve Deep Team Problems

Inbox zero was sold to us by productivity experts, and yet it never seems to happen in real life. When we take a step back, the pattern becomes clear: Fast responses to email -> People learn you get stuff done -> More requests for your help appear in your inbox -> … . Without a fancy label, we engaged in Systems Thinking, a mental model that looks at a problem not in isolation but as a series of interconnected parts. For once, I will find a use for GenAI.

What is Systems Thinking in the Agile World?

Systems Thinking is a mental model that views problems as interconnected parts rather than isolated issues, it means:

• Looking beyond surface symptoms to find deeper causes
• Understanding how changes in one area affect others
• Identifying feedback loops that create recurring problems
• Avoiding local optimization at the expense of the overall system

Systems Thinking in Scrum Master Training

In CSM workshops, my final major exercise is to give attendees a coaching problem and ask them to consider how they would coach through it.

Throughout the workshop, we cover team-level improvements through:

• Retrospectives and clear action items
• Working Agreements to clarify how we work together
• Improving Product Backlog Refinement
• Sprint Backlogs and improving flow
• Definition of Done and improving quality
• Most groups even ask questions that get us to discuss what motivates people
• … the list goes on

Most of these are clear and likely sources of improvement for a good Scrum Master. The final exercise is intended to take them to a higher level, helping them start thinking about the bigger picture and ask why some challenges are recurring problems. The group is given one of several stories about teams that are struggling, and they’re invited to help coach the fictional team. The problems are challenging enough that small improvements at the team level are helpful, but not enough. The idea is to consider many real problems at work and to understand the bigger picture.

My goal is to introduce the attendees to the rudiments of Systems thinking. I give people just a few questions:

• What happened weeks ago? months ago? years ago? Many challenges are buried years in the past.
• If this continues, what will the effects be in months? years? Delay effects are especially interesting as they hide cause and effect.
• Use different perspectives? How would someone with a different background see this situation? How would a manager see it? A senior executive?
• If we looked from 30,000 ft, what would it look like?
• What assumptions are we making?
• …

There isn’t a complete list of questions; I normally suggest attendees look at the Water’s Centre System Thinking Cards¹. However, it’s hard to keep the questions in mind while exploring a problem. I’ve been using systems thinking as a lens or mental model for years, and I still forget to ask some of the questions. For someone new, the whole thing can be intimidating, and that isn’t good.

Using GenAI as a Systems Thinking Coach

I’ve been thinking about how GenAI can aid critical thinking rather than generate more slop. That is when I realized I could create a set of prompts to help with using Systems Thinking as a mental model. I created it as a Claude Skill², so that it can be added to your Claude environment and always be available for use. (The standard has been adopted by OpenAI and others, as Agent Skills so without too much effort, you should be able to use the skill in most LLMs)

Since most people in a workshop don’t already have Claude Code running, they just copied my Claude skill text: https://github.com/mlevison/systems-thinking — specifically this file: references/questions-of-a-systems-thinker.md and put it at the start of their conversations

I also crafted a challenging situation where I thought Systems thinking would help. Let’s run with an example here:

As usual with a GenAI model, the conversation is long. I’m excerpting key details and then sharing my response.

Let me start with a foundational question: What organizational forces or constraints led to the decision to horizontally stratify teams in the first place?

My response: In the past, before Scrum, the whole stratified by specialized teams. Since Scrum has come in, nothing has changed. Everyone says it is just more efficient.

The phrase “everyone says it is just more efficient” is particularly telling—it points to an unexamined mental model that may be creating a self-reinforcing loop. … Question for you: When people say “it’s more efficient,” what evidence are they using? Are they measuring:

• Individual utilization rates? (people staying busy in their specialty)
• Or actual throughput of completed, valuable features to customers?

My response: They’re measuring how much work (output) each person gets done.

There it is. This is a classic systems thinking trap called local optimization at the expense of global optimization. The organization is measuring output (how busy each specialist is) but not outcome (how quickly customer value flows through the system). … What’s happening to the people in this system? You mentioned team members “checking out mentally around hour five.” What are you observing in terms of morale, turnover, or people’s sense of effectiveness?

My response: At the start of each big release people are enthusiastic. As release wears on, priorities change and the pressure to get everything done increases.

My response: They tend to comment on the requirements because that is another department.

Visualizing Problems with Causal Loop Diagrams

This could go on forever, and the conversation was already over 300 lines, so I asked: “Before we go any further, let’s visualize what we have so far with a Causal Loop diagram”. Diagrams in the Markdown world are typically done with Mermaid (what a name), and so I rendered them with the free Mermaid editor

This picture illustrates some of the key challenges in the conversation and would make a good starting point for further conversation.

Conclusion: Using GenAI for Deeper Team Analysis

Be careful of taking the outputs of the model too seriously, remember its only job is to predict the next most likely token, it has no understanding. In addition, if you read the much longer version of the conversation, you will see the model made a few logical leaps that would need to be checked.

Systems Thinking is most helpful when it is part of a conversation among a group working on a problem together. It is a tool that helps build a shared understanding of the world. The intention of this skill is to make it easier to have those conversations and to ask questions that might have been missed. Please use this with others and not in isolation.

Footnotes

1. Waters Centre Systems Thinking Cards: https://thinkingtoolsstudio.waterscenterst.org/cards ↩

2. Claude Skills are specialized instruction sets that teach Claude how to excel at specific tasks - like creating professional documents, building presentations, or analyzing systems. A big benefit is that skills are only loaded into Claude’s working memory when needed. So our Systems Thinking skill sits in the background, and Claude will pull it in when your conversation calls for it. This saves on tokens and makes it’s use more efficient. ↩