I'm generally pretty pro-AI with one major exception: agentic coding. My consistent impression is that agentic coding does not actually improve productivity and deteriorates the user's comfort and familiarity with the codebase. I formed that impression from:
my own personal experiences Every time I use agentic coding tools I'm consistently unimpressed with the quality of the results.
my experiences interviewing candidates I allow interview candidates to use agentic coding tools and candidates who do so consistently performed worse than other candidates, failing to complete the challenge or producing incorrect results 1 . This was a huge surprise to me at first because I expected agentic coding to confer an unfair advantage but … nope!
research studies Studies like the Becker study and Shen study show that users of agentic coding perform no better and sometimes worse when you measure productivity in terms of fixed outcomes rather than code velocity/volume.
I don't believe agentic coding is a lost cause, but I do believe agentic coding in its present incarnation is doing more harm than good to software development. I also believe it is still worthwhile to push on the inadequacies of agentic coding so that it empowers developers and improves code quality.
However, in this post I'm taking a different tack: I want to present other ways to leverage AI for software development. I believe that agentic coding has so captured the cultural imagination that people are sleeping on other good and underexplored solutions to AI-assisted software development.
The master cue
I like to design tools and interfaces from first principles rather than reacting to industry trends/hype and I've accrued quite a few general design principles from over a decade of working in DevProd and also an even longer history of open source projects and contributions.
One of those design principles is my personal "master cue", which is:
A good tool or interface should keep the user in a flow state as long as possible
... continue reading