Why (almost) zero cost software development is going to make us all programmers
Everything I knew about software development no longer applies.
Without using any software development knowledge, I wrote a super sophisticated piece of software this week. It would have been a solid month of work to write it last year. The year before, it would have been a license for an enterprise text generation tool. It took me two minutes to create.
It turns out everything I knew about software development no longer applies, as of the release of ChatGPT.
Today, when I found myself in conversation with a private equity fund about how they might streamline their internal processes through the power of generative AI, I had to shake my time-tested intuitions. They outlined a variety of tasks primarily centered on the conversion of data from one format to another. Immediately, my gut instinct kicked in; it screamed 'prioritize'! I began to mentally sort these activities based on their potential ROI, because that’s what you do with technology projects . . . right?
Historically, the development of software (especially AI models) has been costly and time-consuming. Many seasoned technologists have walked headlong into building projects that don’t get used, driven by the excitement of creating the 'shiny thing.' Through my own experience in making crucial decisions about what to build, I've developed a knack for evaluating risk factors - whether something is valuable, usable, and buildable — before taking the big step of committing engineering resources.
This risk-assessment approach leads to discarding many potential solutions that initially sounded good, because the team quickly learns those solutions simply posed too high a risk. A risk of not being valued by the user. A risk of not being easy to use. A risk of not actually being possible to build.
Occasionally, we'd even stumble upon ultra-simple solutions that required zero fancy tech. Those were places where we could look very silly in hindsight for employing a team of highly paid programmers to employ machine learning.
However, the landscape of AI has shifted dramatically, largely thanks to the introduction of generative AI models as utility services. The cost barrier to getting custom software written and executed has been driven to zero, making them perfect for low-risk experimentation before committing to full-scale development.
In a stunning twist of events, AI has brought the cost of software creation down to virtually zero. You pay a tiny fraction of a cent per word (token) generated by these new models. To paint a clearer picture of what it’s like to write software in 2023, let me walk you through an example. For a recent project, I developed a diagram, akin to a tree, illustrating various possible solutions to pursue. To transform this tree into a long-form business-centric piece, suitable for the business stakeholders, I utilized ChatGPT. After two minutes, I had a well-crafted prompt explaining the context, the expected input format and the desired output, and I was ready to go. Unfortunately, the diagram's entirety couldn't be directly translated into a textual format. Instead, I adapted by copy/pasting the nodes corresponding to each paragraph and let AI work its magic. The results were impressive. It required 10 minutes to finish and saving me about an hour of tedious manual work.
I could create a perfectly tailored routine for my exact task, then discard it, without worrying about reuse. It might seem counter-intuitive, but when creation is virtually free, the value of reuse diminishes.
This mirrors the disruptive paradigm highlighted by Clayton Christensen. No respectable software engineer would claim that what I did was equivalent to enterprise software - it was a laughable (emulation of) scrap of code at best. But it was written without using any knowledge of software engineering, and written instantly. Talk about agile!
What I knew was what I needed done, not how to do it. And no one will know what they need done more than the user of the software.
The world of software, including the complex tasks solved by AI, is no longer a high-cost, high-risk venture; instead, it's an open playground for businesses to experiment, innovate, and drive value before they worry about ‘industrializing’ things into traditional software.