Wondering if I use AI-assisted coding differently than a lot of developers. I don’t chat with AI forever to come up with a perfect plan and then have it execute it. I use AI the same way I would code traditionally, iterating quickly with dozens of quick changes. Still feels like programming.
@manton my mindset is to use it as if pair programming. It just so happens my pair is absurdly fast and doesn’t care if I ask them to redo something, try 3 different approaches in parallel, or go at it alone for a few minutes while I grab some coffee.
@manton I’ve largely been using it the same way. The difference is that the speed with which the LLM can implement my whims means I can quickly try more variations of an idea and consider riskier ones. So I’m making more commits, opening more PRs, testing ideas, and merging only the best ones.
@manton I think Cal Newport said something to the effect that most coders are doing it the way you do.
@unravioli Makes sense! Iteration is so fast that there isn’t much of a cost to doing down dead ends and then coming back to try something new.
@manton I’ve been working on a project that way. Iterating over the design, start with something simple and easy to describe then tweak it up. What’s great about it is that it takes 15 seconds to iterate something that could take a full day to do by hand.
@manton I’m the same. I like the plan mode only initially because it’s good at asking questions whilst exploring the codebase. Then I skim through it and just let it do what it wants and then iterate over that. Very rare for me to spend more than 5 mins for any plan. Even without planning it’s all great these days.
@manton It’s a mix for me mostly based around how much I already know about the domain I’m working in. Given I started in a new role and in a slightly new tech space, there’s a lot of familiarization I need to do. In my previous role, it was much more as you describe since I knew the systems and product well already.
@vincent I should probably spend more time planning, but I love to just dive into it. Depends on the feature, though. For a new app or large feature, planning is good to work out problems early.
