There’s no doubt that coding agents can make an individual developer more productive, at least in the initial stage of actually creating the code. However, I’ve got bad news for all the software development organizations out there. AI isn’t going to make you more productive. To do that, you need to improve your process.
Why is that? Let me explain.
I’m a proponent of the #NoEstimates movement. One of the central premises is that developer speed does not determine when you’ll be done. Your organization’s process does. Instead of relying on estimates you should look at the historical data and then use forecasting to determine when a set of features (i.e. a release), will be done. For more information see the No Estimates book.
This is also pointed out in detail in the appendix of Accelerate book. There they lay out in detail how they statistically established that organizational capabilities, not individual speed, drive lead time. Also there’s a 10-year retrospective that shows more data on this: Accelerate State of DevOps Report 2024
Yes, your developers are probably cranking out new code faster than ever, but, they’re being slowed down by more and more code needing to be reviewed, and bugs needed to be fixed because they fell through review because of that onslaught of code. All that new code is still sitting there for days waiting to be reviewed just like it was when it was all hand coded, especially when the pull request is large, which I’m guessing it common when generated by coding agents.
So, it doesn’t matter how fast your developers crank out code, whether by hand, or with coding agents. It will still take the same amount of time it usually takes for your organization. You’ll need to improve your process. Maybe look into trunk-based development, automating PRs in some way (like faster unit tests, linting, static analysis tools, and the like). Look into continuous integration and deployment. Some more upfront planning might help, like involving developers in feature design rather than just handing a design over the wall to them, where they then have to investigate and fill in the gaps product managers and designers didn’t think about.
It’s not easy and people being people, you probably can’t go much faster. It will take a lot of work, patience, and measurement.