I consider myself a seasoned programmer, but I wanted to see how much could be done with free AI tools. So I decided to play with ChatGPT and see what I could create by tying my hands and not coding at all. It was an interesting experiment in herding a cat with memory loss, but it came out interesting. I found it hard not to intervene and give the code or answer when errors or bugs arose. It would have been MUCH quicker just to code it myself, but it was fun to see the future as an AI herder, directing development vs coding it. I enjoyed working with the AI chat agent, and overall it was fun. All assets were AI-created, including images, text, and SVG code. Feel free to PLAY the game and check out the code we made. It will only work on a desktop or tablet; it’s a bit too big for most phones. It is 2-player, and you can click the buttons or use the keyboard!
Play Snozzie vs. The Crimson Rascal!
So, did you play it long enough to see its funny quirks?
Reflections on the process:
It took a lot of prompting and pushing to get the game where I wanted it to be.
Like many others, I have been thinking a lot about a future with AI. Anything we say will surely not age well, and most “AI” now really is an advanced search parlor trick, a handy one, but not true AI, yet. Coders will be AI herders in the future. The human in the process is the director, and they have the vision of the solution or goal. Like a good manager, they lead the flock down the right path and direct them when they hit trouble. I have also made a few PowerShell scripts, and I made a point not to code to see what I could get. It struggled with solving the details or bugs and understanding more complex goals or plans. I purposely used the free lower-end chat GPT 3.5 to test the worst-case results; other newer or more optimized tools would be better. I eventually got the hang of wrangling and herding it, not unlike the early days of Google. Where it required a new skill set to find what you were looking for, and if you mastered it, you had unlocked new information. I could have coded or fixed the bugs much faster on my own. I may share the hilarious attempts made repeatedly, but ultimately, I got usable code.
It was not unlike managing green, freshman student workers on a project. They say, “Oh yeah, I know X!” but when the rubber hits the road, they need more guidance to flourish and find the solution.
I am at an awkward point in my career. I am transitioning more and more from technical roles into management and leadership ones. So, herding and wrangling AI have felt strangely related. The skills I use to manage a project or team apply more than I expected.
Tips:
A few basic concepts will help you in your own AI herding. I am sure these will be outdated and embarrassing to reread in a few years, lol.
-Work in chunks or functions. Only work on one feature or issue at a time.
-Read the code it makes, and if you don’t understand why or how something is done, ask to have it explained. You will either learn or correct a bug.
-Do not blindly trust the code or answers you get.
-But do use it to check your work and find mistakes.
-Be polite, nice, and speak respectfully to the chat agent. Like a real team member, it will affect engagement and code quality. (I tested this quite a bit)
-Lean into what it does well and avoid what it does not. I found it excellent and was handy in doing tedious repetitive work, like upgrading code to PHP 8, but struggled when I needed solutions for complex problems.