My AI learning journey has been long, and littered with gaps. I first got really interested in AI and machine learning back in 2018 and joined a team whose product used some machine learning. Since then I’ve tried to keep up with the GenAI/LLM space as it exponentially grew. Workshops, webinars, 30 days of Ai with Ministry of Testing, books, blog posts, talks, and even co-facilitating workshops myself helped.
Yet, the one thing I hadn’t done is to use AI assistance tools hands-on for software development. I retired from full time work in 2022. Having paired and ensembled for so much of my career, I find I can’t get motivated to re-learn coding skills and create apps on my own. I helped a bit with open source products, but that was before everyone started using AI assistants like GitHub CoPilot and Claude Code.
Getting past the blockage
I’ve continued to seize learning opportunities when I can. And, it seems that the GenAI/LLM landscape changes every day. Also, when did it become ok to call it “AI”, which what we are using really is not? Things moved way too fast for me. I fell way behind the technology times. It felt like there was a huge hurdle I could just not get over.
Pairing to learn, one-on-one
Feeling overwhelmed, I decided I needed to pay someone to teach me how to build something with an LLM. Fortunately, Katja Obring, who’s one of the leading lights of GenAI/LLMs in the quality engineering space, had some availability. We arranged to have six one-hour learning sessions together. I made her charge me for them, I knew it was a superb investment on my part. And I got a great return on that investment.
Kat created an excellent learning experience for me. We started with me getting a great overview of LLM tools and the usage possibilities they bring. Not only coding assistants, but things like NotebookLM, which I had heard of but not tried (and it has proven useful!)
We discussed what to build together for my learning. We settled on putting together this capability: Download a Google doc with a blog post draft, convert it to markdown, and post it as a draft to my WordPress blog. Together, we used a mind map to plan the steps needed in this process.
Encouraging benefits
One of my biggest takeaways here is that GenAI can be whatever tool you want it to be. I can look at my workflow. What would I like to do myself? What do I want AI to do? Hearing how Kat uses these tools helped me get a feel for my own possible use cases.
As far as the tool to use, we settled quickly on using Claude code with Visual Studio Code. My IDE and terminal skills had gotten quite rusty as I haven’t really used any of that for a couple of years. Kat patiently guided me. (“HOW do I open a terminal window? Oh, duh.”) She reminded me to do a brew update claude-code at the start of each session. Being reminded of good practices is as valuable as learning new skills!
We explored various approaches with Claude Code to make the post-a-draft-on-WordPress and ended up creating a skill file. Kat explained that a skill can provide more control and gives better ideas of what it can do than an agent.
Learning how to work with Claude was interesting. It took me a bit to realize how easy it is to ask it questions or just give it a screenshot with an error message and get help. My biggest struggle was in the authentication parts to let my new tool access my Google drive, and subsequently, my WordPress blog. I was proud that I figured some of that out on my own – with help from Claude.
Feeling freed up
The thing I built isn’t important. What I needed was the practice of using the tools and getting to pair with a human to get a good idea of how practitioners use this technology. Kat gave me some good suggestions for continuing my learning.
A week after my last session with Kat, I joined a free online workshop with Woody Zuill, called “Build together, learn together”. In the workshop, we worked in small groups, combining software teaming / mobbing / ensembling with using AI assistance tools. Thanks to everything I learned with Kat, I was able to get really engaged with the activities. I was too timid to offer to be the “driver” with my own Claude Code. Another group member used the Claude Code CLI as we planned and had Claude assist us in creating an online Lean Coffee application. I got to experience how real life teams work in our new age of AI assistance.
And, I decided to try building a Lean Coffee app myself with my own Claude Code. I had to install some extra things (Ollama app, for one), but Claude guided me through that. In a few minutes, I had a rudimentary Lean Coffee app running on my localhost.
Going forward
I don’t see myself building my own apps. Coding solo, even with an LLM, is not my jam. I do want to continue learning, starting with the resources Kat and others have shared with me.
There’s a valuable takeaway for me from my sessions with Kat and subsequently, the workshop with Woody that my new, if
basic, knowledge and skills enabled me to enjoy. I now see how teams can work with these tools, how it changes the way people develop software. I can see how teams using good practices, like working in small increments iteratively, can get huge benefits by letting AI tools do the grunt work. Hopefully I will find it easier and faster to keep learning, because I’ve grokked a big picture of how this all can work.
At the very least, I can listen to others’ experiences and follow along! Hopefully, I will find opportunities to pair and ensemble with people using these tools. If you want someone to pair with you to build a thing, please pick me! Seriously!
