So, many of us have been working with various IDE which has tightly integrated conversation and code generation features. With minimal prompting, one can extend existing programs to add intricate and complex functionality. It writes good code, and matches all the surrounding style etc.<p>Traditionally, programming (in a meta sense) is about this loop between thinking about what to code, and then writing that code. Sometimes you have to break the loop and go learn how to do something.<p>Very very generically speaking, your work-a-day software engineer might spend a morning like this:<p>A. Spend an hour reviewing requirements and decide what code to write, how it should be factored, etc.<p>B. Spend a few hours writing that code<p>Now, with good code generation from AI, step B might be cut down to 30 minutes for time to generate and then review the generated code.<p>I'm finding this experience to be very interesting.<p>Do you think this is going to be widespread enough that expectations around productivity are going to raise? Is it more likely that I have to deliver 4x more jira points or that I get 4x more free time each data in the office?
$20 pays for about an hour of extra productivity per month. If you get that much, you have your ROI.<p>Is that an hour not spent googling? Is that an hour of extra energy because I didn't need to sift through a 5000 line log dump? It's pretty good.<p>The actual value I get from agents is it's a rubber duck+Google, not that it writes code. It's also a journal of my previous train of thought. I can 'code' while in the train now, or 15 min before a meeting instead of blocking out 4 hours.<p>It's closer to 2x for me, realistically. But anything more than 2 hours per month seems greedy.<p><i>But</i> that 2x productivity doesn't mean I get twice as much done. It means some of my colleagues get laid off, some interview/freelance/side hustle because they're worried about getting replaced by AI. The whole mess means return to office which ends up in 0.5x productivity and 1x net productivity.<p>But we need a few less engineers and managers so I guess that $20 and the $20000 office space is worth it in the end.
I wouldn't be surprised if expectations have already increased in many years because of these tools. A lot of coding tests for new roles seem to expect a level of work that would have taken a week or so a few years back, and I half suspect they've been tuned on applicants using ChatGPT/Claude/whatever to handle everything.<p>So it would make sense for a lot of companies to think "well, these LLMs making coding 'trivial' now, let's ask our developers to do a lot more work now they don't need to think too much about it". The fact that debugging AI generated code can be hellish, or that these systems have more and more obvious limitations the further you get from the ideal situation (coding CRUD systems in JavaScript or PHP) probably won't cross their minds.
slowly over time yes but don't worry, you will always just have to do what an average human does. If you want a raise, a little bit more than average. If you want to be a founder? 100x more. It's the same system, same median averages and bell curve.
> Traditionally, programming (in a meta sense) is about this loop between thinking about what to code, and then writing that code. Sometimes you have to break the loop and go learn how to do something.<p>Nope. Meta-traditionally-philosophically programming is about software _engineering_.<p>Code is just a byproduct, intermediary, leftovers etc of the process. Liability, not an asset.