It's been a while since I integrated AI agents into my programming workflow. At work I mainly use Claude Code. For private projects I'm a cheapskate, which means I use whatever is the cheapest (which at the time of this writing happens to be Gemini, given their very generous free plan).
After about a year of using them almost on a daily basis, I've noticed my AI usage falls into two (generally) distinct camps.
Camp 1: Productivity Gains
The first camp is what I'd call productivity gains. This is when I offload the type of work that's necessary but is either rote work or is somewhat at the edge of my domain knowledge.
For instance, if I'm working on a fullstack feature and want to make the UI look good (building UIs not being my core skill), I may use an AI agent to take a first pass at the implementation. This saves me a few rounds of Google searches, copy/pasting suggestions from StackOverflow, etc.
Although implementing good looking UIs may not be my core skill, I'm still fairly confident at distinguishing a good implementation from a bad one, which means I have a very good idea of what I'm looking for. Using AI agents to help me with such problems saves me time and increases the scope of what I'm capable of building, in turn making me more productive.
Camp 2: Good Lazy
The second (more interesting) camp is when I'm being plain lazy, but of the good kind!
A good example here is creating pull requests on Github. At work, I wrote a custom
slash command for Claude Code to create PRs on Github. Instructions for this command
ask Claude to take a look at the diff between the main development branch and the
current branch, and create a PR on Github using the gh CLI, assigning it a title and
description based on the git diff and formatting the description based on the PR
template checked in to the git repository.
Of course, I could easily do all those things myself, and the PR's description would be a bit more focused (LLMs tend to blabber a lot!), having a custom command for this saves me time. I'm being lazy, but I'd argue that this is the good lazy.
At the end of the day, I feel that when using AI agents, it's very important to know the difference between genuine productivity gains and shortcuts to complacency. While the two of them are not mutually exclusive, comfort that leads to complacency isn't something desirable in the long run. Hopefully this quick reflection of mine encourages you to be mindful of your own workflows with AI agents!