The future of coding will just be telling a computer what to do in plain English (or whatever language you speak). Prompting will become the new coding, and hand coding will only only be used to intervene when the AI fails.
We’re currently at the stage in the AI cycle where AI is becoming pretty good at answering questions and following instructions (eg. format this document into JSON), but we’re not yet at the stage where the AI actually executes on those answers. AI currently gives you your text response (or an image/video if it’s generative AI), but then you have to do something with that response (eg. copy & pasting the AI-generated code into your IDE).
Once AI is able to execute (eg. when you ask it to build an application, it actually builds the application instead of just returning the code), AI will really become like a superpower. The amount of human labor required at that point will drastically decrease, along with the cost of building and maintaining applications, while applications themselves get commoditized since at that point your grandma can build one.
It would seem that the AI execution layer should be relatively simple to implement.
Stages of AI
- Text response and media generation (eg. ChatGPT, StableDiffusion)
- Execution (eg. AutoGPT, Codex)
Hand coding becoming obsolete
Currently most coding looks like this:
- Code until you need help
- Either Google search → Stackoverflow, sort through SEO-optimized blogspam to find a random tutorial that answers your question, etc. Or look at documentation. Or just ask an AI like ChatGPT, Bing Chat, you.com, etc.
- Copy & paste code
With AI able to execute, that gets reduced to:
- Tell AI what you want in a prompt
- If it’s not working correctly, then follow up with another prompt, or go “under the hood” and edit the code / do the thing yourself
So instead of spending a month building some web application, now you just tell it what you want in a prompt, and so long as the AI works, your work is finished.
Want to make modifications to the app? Just write a prompt, no need to fire up an IDE and sift through some complicated codebase. You don’t even need to write the prompt, just tell your phone by voice and your AI voice assistant will relay the request.
Now obviously it’s going to take time to get to a stage where the entire application of a non-trivial complexity can be built in a single prompt, but there’s no denying that we’re dramatically moving up the abstraction ladder to a point where any non-technical person will be able to build anything they want.
Programmers online often protest this by saying “well AI on its own won’t be able to solve this special complicated edge case”. Sure, I don’t think anyone’s claiming that all hand programming will completely vanish (after all there are still people programming in C). There will always be exceptions. But the vast majority of tech work is glorified plumbing, not inventing some novel algorithm, data structure, or systems design architecture.
The average digital plumber, err programmer, is therefore f*cked. But if it’s any solace, so is everyone else – designers, data analysts, etc. Hence the obvious need for a universal basic income, but seems that won’t be implemented until there’s enough suffering and the writing on the wall is so clear that it’s politically viable.
Remember that people are incentivized to want their jobs to continue to exist because in our pre-UBI capitalist society, most peoples’ jobs are their livelihoods. They’ll say “no this new AI thing won’t work, you still need to hire a human like me to do the job”. Sure maybe for now, but at the pace AI is developing, your work will be automated sooner than you probably think.
J. Savage 