Can AI Replace a Scrum Master?

Some friends were discussing a post by Jeff Sutherland about how AI will make estimation easy. What I thought was odd about that post was looking at where AI would begin to replace various Scrum activities, so I thought I’d plug in a prompt to see what AI will cook up for something like a sprint planning. You can see the results below.

Here’s a script that a new Scrum Master can use during sprint planning to help teams estimate work items, consistent with the Scrum Guide:


Scrum Master Script for Sprint Planning

Introduction: “Welcome everyone to our sprint planning meeting. Today, we will focus on defining the work for the upcoming sprint and estimating the effort required for each item. Our goal is to create a clear plan that aligns with our team capacity and ensures we deliver value.”

1. Review Sprint Goal: “Before we dive into the backlog, let’s discuss our sprint goal. What is the primary objective we want to achieve in this sprint? This will guide our prioritization of work items.”

2. Present the Backlog: “Let’s take a look at the product backlog. We will go through the items that are prioritized for this sprint. [Refer to the backlog.] Please remember, we only pick items that align with our sprint goal.”

3. Discuss Each Item: “I’ll read out each item, and I’d like us to discuss it briefly. Can we clarify any uncertainties about this item? [Pause for discussion.]”

4. Estimation Technique: “For estimation, we will use [choose an estimation technique like Planning Poker]. This helps us reach a consensus on the effort required for each item. Remember, we’re estimating relative effort rather than exact time.”

5. Estimation Process:

  • “Let’s take the first item. Everyone, please take a moment to consider your estimate based on your understanding of the work involved.”
  • “On my count, we’ll all reveal our estimates at the same time.”
  • “If there’s a significant difference in estimates, let’s discuss those perspectives to understand why.”

6. Document Estimates: “I’ll record the estimates as we go along. This will help us track our progress and adjust future planning if necessary.”

7. Assess Capacity: “Now, let’s evaluate our team’s capacity for this sprint. Based on our estimates and historical velocity, how many story points can we realistically commit to?”

8. Finalizing the Sprint Backlog: “Based on our capacity and the estimates, let’s finalize our sprint backlog. Are there any items that we think should be removed or added? Let’s ensure we have a balanced workload.”

9. Commitment: “Once we’ve agreed on our sprint backlog, I’d like each of you to commit to the items you’re comfortable with. Remember, commitment means working together to achieve these goals.”

10. Wrap Up: “Great job today! I appreciate everyone’s input and collaboration. Let’s ensure we communicate effectively throughout the sprint and keep each other updated on our progress.”

Well, what do you think?

Before doing this, I assumed AI would reach about 80% of what would be appropriate, and I’d say that’s what we have above. Now, some folks will want to argue with what I’m about to poke at, but I want to clarify that this exercise aims to see if AI can replace a scrum master.

Step 3

I don’t have any specific feedback until we get to item 3, which is to discuss each item. That means the Introduction, steps 1 and 2, are fine to the point I don’t feel it’s worth debating anything. In point 3, however, something interesting happens. The script has the Scrum Master read each item out loud to start a discussion. There isn’t obviously anything wrong with the Scrum Master reading things to the group; it’s one of many ways to discuss backlog items. In the long term, if the Scrum Master reads everything to the team, the team will likely become dependent on the Scrum Master and others and will treat the backlogs as someone else’s problem. This is a small thing, but it’s the first point I see that stands out.

Estimation

The script mentions they’ll estimate using a preferred technique on relative effort. This type of advice would fit with what you see all over the industry, but here’s where my criticism gets a bit more pointed. The Scrum Guide does ask that the sizing of items happen by the developers, but makes no statement about what that size looks like. Time-based estimates are allowed, as is right-sizing.

Commitment

The last two I want to discuss are points 7 and 9. Again, this advice is consistent across many Scrum Masters and materials. We see the shift from vague relative effort estimation to using story points. Again, this is the most common choice, but it is one of many. I’d like to draw attention to the last statement about committing. Step 9 echoes the importance of commitment.

There are no sprint commitments in Scrum!

This is where we see some of the reality of how well AI does. It did a good job replicating what many people do, but the problem is that it’s replicating something that hasn’t been in Scrum since 2011.

The only thing the team commits to in a sprint is the goal. All the rest of this about committing to the individual items and content of their sprint is a thing of the past that should stay there.

How’d AI Do?

So, where does that leave me in my guess that AI would get us 80% there? It did that almost exactly. There were 10 or so points, and 2 of them, I would say, were faulty. What’s tricky here, and this is what people are realizing, is that while 80% might be fine, you don’t know where the problematic 20% is. It still takes an expert to alter, adjust, and edit AI results.

Having said that, this AI produced a legitimate script that many Scrum Masters use. Sadly, most Scrum Masters who use a script like this create sub-optimal results, frustrate teams, and get their names on the list for the next round of layoffs.