Beast

 

At some point in the early 90s I tried rebuilding a game I loved: Beast. It was written procedurally in Basic over couple of months of on and off coding. Clunky, filled with bugs but I felt proud.

About 25 years later, with 12 years of professional experience, I tried again. In about an hour, I had a working version. No bugs this time. Of course, Beast is not a complicated game but by going object-oriented with multi-threading it was even easier.

And then, another 8 years later, I built Beast again. In record time: 25 minutes using with 7 non-technical prompts. No logic or technical guidance - just plain-english rules.

We're well past the point of arguing AI's abilities. Instead, I wondered: In those 25 minutes, what parts of my own experience did I leverage?

1. Strategic Iterations: Even if the LLMs handles complex specs, doing things all at once will still overwhelm the human. There are too many things to track at once. The ability to break things down into discrete, logical, and verifiable units is an important skill. It's Test-Driven Prompting.
   
   
2. Technically Feasible: Even though I didn't give technical instructions, my requirements were still grounded in and optimized for the tech I was using. If you've ever done freelancing/consulting work, you've fielded outlandishly impossible client requests. As a freelancer, you learn to say 'No.' Without knowing the rules and the nuance, it can be a wild goose chase.
   
   
3. Problem Solving: When results were wrong, I knew why. Without looking at the code, I could visualize the execution and could make targeted prompts to fix issues quickly. It's like the parable of the experienced mechanic's invoice: labor costs $1, access to their expertise? $9,999.
   
   
4. Balance: Some improvements can be made faster than writing a prompt. Knowing when to use AI vs doing things manually (along with the ability to do them manually) is important. You don't need a jetpack to walk down to the mailbox.
   
   
5. Scope Creep: Just because you can, doesn't mean you should. AI is like a monkey's paw. For better or worse, if you ask, it will (try to) deliver. It's an invitation for bloat. Coding agents will get better and maybe push back but there's a limit to what we should offload.

All of these are important programming & design principles for building good software. AI can do the bulk of the work yet I keep running into the importance of being a skilled and knowledgeable operator. Anyone can use a hammer, a saw, and a drill - but that doesn't mean everyone's out there building houses. The principles behind building resilient and maintainable code bases, and the steps to go from proof of concept to enterprise software will matter greatly.

Which leads me to think: If I were 8 years old today, I'd probably still start the same way.