What will ChatGPT mean for a mediocre programmer like me?

It seems unclear to me which level of expertise will be most affected by AI becoming a defacto tool in the software development process. Most people are assuming senior/proficient devs will still be needed to spot errors, and junior ones less so...

But you could also argue AI code generation could be a great equalizer and mediocre programmers could be as effective, if not more effective than more skilled programmers since the skillset will change from the ability to write good code, to the ability to understand specs, and use the tooling.

So start learning how to use GPT in your workflow and maybe you'll be pleasantly surprised by where all of this leads!

There is so much more to creating a successful software system than writing code. Getting your code to work is, in my experience, about 10-20% of the effort involved in making software effective IRL. And it's by far the easiest 10%. Data modeling, architecture, performance tuning, tech decisions, design, load testing, running on a novel infrastructure, scalability, migrations, operations, security, a process for adding new features and fixing bugs, integration with other systems, etc.. tend to make up the vast majority of the work in a real world software project.

These aspects of programming are relatively impervious to AI takeover because a) they require clear intentions (i.e. a group of humans must collectively agree on the business outcomes they are shooting for) and b) they require constant collection of and tuning against empirical data (novel data that is absent from the AI training sets by definition).

If your job is solely to write code according to someone's prompt, AI will replace your job. But the person deciding what that prompt should be in the first place... that's the value proposition that AI cannot touch.

AIs are tools. The successful programmers will be the ones that can best use the tool towards their advantage. AIs can't function on their own. They need to be guided towards a solution for a human problem. That's where the new breed of programmers will succeed. You tell it what you want. It returns a result. Your job is to make sure it fills the need. I would even say that you are at the same level as everyone else since mediocre or award winning programmers are facing the same issue. Giving up because you think it will do your job and you don't have a role to play is not the answer.

You won't understand why the generated code you use that still works is suboptimal, and you'll internalize antipatterns that still work nonetheless. And your skill ceiling and ability to work in mature codebases will scale linearly with the token context limit.