Leaning Into Generative AI Development
Posted by Josh Sherman on in Software Development
While I've been no stranger to using AI to help me with my coding tasks, recently I've really leaned into generative AI development. I'm not talking about chatting through a problem with ChatGPT, or playing the copy paste game to debug something. I'm talking about taking my hands off the wheel like Tyler Durden, and letting the robots decide my fate.
Okay, so we're not driving off the highway, and I have to do a lot of steering, but you get the idea.
It's a new era, there will be casualties
I've been doing this computer thing for my entire career. I wrote my first BASIC app in the 1980s. I got to experience the Internet in its infancy. I've seen a lot of dinosaurs continue to lead productive careers.
This technology wave is nothing like that. The trends I've been through have all seemingly expanded the market. Now I'm seeing people downsized because they aren't adopting new tools and refining their workflows around generative AI.
Some companies are choosing to do more and move faster by combining their existing headcount with new tools and workflows. More often, I'm seeing companies reducing headcount and still being able to move the needle.
And this is just the beginning, the tools today aren't what they were a year ago. Hell, seems like every quarter there's a new wave of improvements that's making it hard to tolerate a team that's scared of change.
The reluctance to evolve in this landscape is already brutal, and I can't imagine it's going to get better.
Rethinking what it means to be a hacker
I got into software development because I absolutely fucking love it.
Telling the computer what to do, there's nothing finer to me. I am always hacking on something. When I'm not coding, I'm probably. coding.
For me, coding isn't necessarily about typing a series of conditionals, it's about creating something out of nothing. It's inherently human to strive to be like the gods, and there's nothing more God-like than creation.
That's not to say I'm a god, but the act of creation is exciting to me. My first BASIC program in the 1980s spat my name, "JOSH" across the screen over and over again.
And in that moment, I swear I was infinite.
The first of many runaway loops throughout my career.
Generative AI development doesn't change a thing for me. The point is to build something, not form blisters from typing if statements over and over. While far from perfect, gen AI coding allows me to accomplish this very goal, faster.
Done is still better than perfect
I've said this before, nobody wins any awards for beautiful code.
Code depreciates faster than a car being driven off the lot. We throw so much of it away. Projects get abandoned all the time. Buggy code gets rewritten. Frameworks are moving so damn fast that if you wait a year, you're multiple major versions behind.
Given everything we know, why do we insist that the robots write this subjectively perfect code?
Yes, code generated via an LLM is far from perfect. Guess what folks, these robots are trained on our shit code, so why would you ever expect perfection?
Not just that, but show me a project that doesn't have a backlog of bugs. Features always take precedence. This isn't new.
Agentic coding in practice
Let the record show that I abhor the term "vibing", but yes, I vibed VPS Showdown. I vibed the living heck out of it.
I don't think you can figure out what's truly possible by following a couple of tutorials. You can't go into something aimless and hope to have an "AHA" moment if ultimately you'll just throw the project away. You have to set out to do something you could have done yourself, but in a new way, to see what's possible and gauge if it's better.
I still made all of the decisions on the stack and the direction to take things. It's far from perfect, but I now have a system built that's in a great position to further refine and scale.
The most important thing is, it's live. It's out in the world and yes, robots helped me build it.
This was accomplished in a matter of weeks during nights and weekends. While totally anecdotal, I know it would have taken me longer if I wasn't bossing a robot around.
And how does this make you feel?
The process was fulfilling, because I accomplished my goal of building this system that I've had kicking around in my head for years. It's fun because I'm learning how to work differently, so many brain wrinkles right now.
It was important because it proved the model for me. I can boss the computer around in a conversational way and build awesome things. I'm proud of the project, even if I didn't type out a bajillion if statements myself.
At least with where we're at right now, the process still required a lot of the expertise I've acquired over the last quarter century. Even this is exciting, because the expertise lives rent free, and I can focus on how I can work with these tools, and perfect new workflows, instead of just… typing in if statements.
This is just the start
While I feel like I've been using conversational AIs forever and a day now, everything is all still very new.
This is exciting and yes, it's scary. If you're not at least a little scared, you're not moving fast enough.
In the coming weeks, you'll be seeing more from me on the topic of generative AI development. I'm going to be diving deeper into the tools I'm using, my workflows, and all of that good stuff.
Not just the "good stuff" either. There's a lot to go wrong. There's a lot that goes wrong. We're so damn early right now that there's no place to go but up.