Thoughts on AI and the Future of Agile Coaching

Is this the tomorrow you worried about yesterday?

(This post follows on the thoughts I expressed in my "The End of Your Knowledge Worker Profession Approaches" post.)

If you're expecting the decades old model of coaching in the technical and Agile space to continue, the impact of AI on your coaching career can be distilled to one word: Grim. For those who understand the model will change - somewhere between "substantial" and "extreme" - and are actively adapting to how things are changing, the future could be bright indeed.

A little background for how I got to where I am today in thinking about AI's impact on Agile coaching.

I continue to play around with ChatGPT. In one case, I used the service to generate text I actually needed and didn't want to create from scratch: A data and privacy policy for StoicAgilist.com. What ChatGPT created was reasonably good. It needed a few edits, but saved me from having to read through dozens of examples that kinda sorta fit my needs or from hiring a lawyer that would charge $50/character and still get it wrong.

I also pressed the AI monster with a few more elaborate coding tasks. For example, I had a need to monitor the router in my home for specific changes. so I asked ChatGPT to…

"Write a Python script that will determine the WAN IP address of a <my brand> router."

What it generated was garbage. The form was fine, but it choked early and often. A few minutes of fiddling with the code revealed it was going to take more time to fix ChatGPT's output than starting from scratch and still not result in a solution that met my needs or standards. I did eventually code the solution I wanted and did so by good old fashioned experimentation and reading the docs.

But I did find that I started using ChatGPT in a different way. Not intentionally, but as a natural consequence of having experienced more of ChatGPT's capabilities. If I needed an example of how to code something specific or, even better, sort out a Python error that didn't refer back to a specific line of my code, ChatGPT quickly gave me what I needed. It had a bigger impact on my debugging effort than it had on my coding effort. (Your mileage, of course, may vary.) What I had in the room as I coded was my experience, the documentation for the tools I was using, and ChatGPT. What I didn't have was Google or Stack Overflow.

Several weeks later, this article appeared on Similarweb's blog: "Stack Overflow is ChatGPT Casualty: Traffic Down 14% in March" Good to know I'm still as smart as the average bear. Of course, there are undoubtedly other factors contributing to the 14% drop. Lots of tech workers laid off in the last several years, for example, who no longer have need of Stack Overflow.

Stack Overflow was rarely - actually, never the source for complex or complete solutions. It was never intended to be. It was (is?), however, an excellent source for solving tiny vexing bits and pieces of specialized, complex, or complete solutions. Perhaps more importantly, it was and still is an excellent platform for incremental learning and competence building. I'll throw in the added benefit of being the source of new information related to products, trends in technology, and coding practices.

Software coding job security isn't what concerns me. I think that's going to be a fading profession regardless the direction AI takes. Likely replaced by the evolutionary successor to "prompt engineers." It's a trajectory I saw emerging more than 15 years ago and was one of several indicators that prompted me to move from software developer/engineer to technical management and eventually Agile coaching.

Which gets me to today and how AI may have an even bigger impact on Agile coaching than it will have on coding.

Several weeks ago, I attended a presentation by Tim Robinson, "Chat-GPT4 for Agility," hosted by Access Agile Freestival (sorry, they lost the recording so no link.) What he's done with AI and Agile is impressive, particularly since he describes himself as a non-programmer. In a nutshell, Robinson converted Steve Jobs' product announcement for the iPhone to text, fed it to ChatGPT-4, and asked it to build out a product backlog with the necessary stories to make the product. As a reverse engineering exercise, this was quite impressive in it's own right. Taking if further, apparently Robinson has worked out a way to drop the AI generated stories into Atlassian's Jira.

Boom. Take Job's customer story and convert it into actionable technical Agile stories in...what...an hour? There is a significant caveat to be made here. Jobs was describing something that actually existed and his presentation had been crafted, polished, and practiced to be extra smooth and shiny. It's highly unlikely any product begins with this level of definition. But perhaps this is something else that will change. With AI's help, is it possible the practical and doable can be more easily separated from fantasy and far fetched? With AI's help, can a product owner more easily split out releases?

I haven’t had the opportunity to review the AI generated product backlog from Robinson's example. But my experience with ChatGPT is such I can well imagine they were far from perfect. I can also imagine they are a lot more polished than many of the backlogs I've seen created over the course of many months. It's this "many months" part of the work where AI will have the greatest influence. As an Agile Coach working with a new team or with teams on new projects, much of my work in the first several months is spent coaching people how to convert the narrative-style stories from stakeholders to the more detailed Agile stories around which the product owners and teams will coordinate their work. This involves working out acceptance criteria, definitions of done, dependencies, and estimating the level of effort needed to complete the work described on any particular Agile story card.

Teaching this is time consuming. It involves a lot of effort to develop skills unrelated to anything Agile or technology. Highly intelligent and tribal technical types generally don't have the soft skills required to work with non-technical stakeholders. They are happiest hobnobbing with their peers or, even better, plying their trade via keyboard and face-to-screen with a computer.

The capabilities of AI have been demonstrated to completely circumvent the need to develop the skills needed to collaborate with stakeholders or, in the case of (non-code) writers, an audience. By using AI to accelerate the delivery of detailed instructions technical types can readily digest, product owners and their teams can quickly get to the business of working out the nitty gritties. No need for some expensive coach to help teach the skills needed to craft actionable user stories. A competent product owner working with AI will do just fine. I can also see a clear path to AI's eventual participation in developing acceptance criteria, definitions of done, work effort estimates, and more. Over time, less for a product owner to do in this space, too.

Of course, a very real downside to all this is that individuals on the teams will not benefit from the development of skills related to communication and collaboration beyond their technical peers. This isn't a good thing. I know this from direct experience. I designed the card that card carrying introverts carry. Breaking out of this shell not only greatly enriched my personal life, it made me a much better software coder, designer, and engineer. It also enabled me to move into management and coaching without becoming yet another data point in the Peter Principle set.

My spidy sense tells me the roles of product owner and Agile coach are going to change significantly in the coming year or two. (Scrum masters, too. But perhaps not as much and not in the same way.) This is the foggy and frayed edge of my thinking on the subject. I've a lot of ideas and speculation. It could be today's "prompt engineer" role folds into that of the product owner. Or the role of the Agile coach folds more into the more situation specific needs of Scrum mastering. Or maybe these roles disappear altogether, replace by something entirely different designed to fill an as-yet undiscovered gap that AI can't replace.

It will be interesting to see if opinions like this...

...end up being worth the paper they're printed on. There are plenty of opinions like this floating about and they read more like snarky hubris, denial, and entitled immunity than cogent prognostication. In my view, they completely disregard the squishy middle between client and programmer (or Agile delivery team, more likely.) Almost as a rule, clients have never been good at accurately describing what they want and I don't see that changing. An equally valid rule is programmers have very poor skills for working with clients and discovering what they want. But they are quick to blame the client when the client complains about the final product. ("Hey, I built exactly what you told me to build.") It could be argued that the raison d'être for the product owner role is to facilitate the process of wrestling with the client to extract what the client wants and then wrestling with the Agile teams to be sure they understand what the client wants. Aligning hallucinations is a thankless and underappreciated part of the product owner's job.

What will change is the communication path between the client and the Agile team. Someone who is skilled at translating the narrative from the client into the instructions needed by an increasingly capable AI could very well generate the Agile stories needed for a high quality product backlog in significantly less time. Couple this with an increasing capable AI for writing and debugging code and...well, time will tell just how safe software programming is as a profession and to what degree. Will whatever evolves from this unholy collaboration even qualify as coding? I'm biased on this score, having earned my coding chops on Assembly, C, and 256K of RAM.

One thing I'm sure of for my own self. I've seen a lot of fads come and go over the past 40+ years. For most of them, I yawned when they took the lime light with drama and flash and barely noticed when they faded just as fast. A few waves, however, grabbed my attention and held it for good reason. They really did represent a significant change - the World Wide Web and bio-engineering are a few examples. The coming of age for AI and tools like ChatGPT, DALL-E, and Copilot very much have the same feel as the other significant waves I've mentioned. I could be wrong, but my plan is to get on my board and ride this wave because the undertow we're witnessing real-time looks vicious.

Shaka, baby!

---

Photo by Jeremy Bishop on Unsplash