After talking to more than 30 developers, executives, analysts, and researchers, MIT Technology Review found that the people building software do not agree on whether AI coding tools make them faster.
What they do agree on is narrower and more useful. The developers interviewed consistently named the same places where the tools help: boilerplate code, writing tests, fixing bugs, and explaining unfamiliar code to someone new to it. Several said the biggest value is getting past the blank page.
The piece also cites a study by Model Evaluation and Threat Research (METR) in which experienced developers believed AI made them 20% faster while measured results showed them 19% slower. I think that gap between felt speed and measured speed is the whole story. Feeling faster is real, and it is not the same thing as shipping faster.
My view, after years of running front-end teams, is that a team gets more from these tools by starting where everyone agrees they work and measuring from there, rather than expecting them to speed up everything at once. Tests and boilerplate first. Architecture later, if ever.
Where has AI coding help held up for your team, and where has it not?
Photo source: https://photos.robertstowe.com/venice