What if AI was coming "everywhere, except in our productivity metrics"? What we're seeing feels like a modern replay of Solow's paradox.
On one hand, AI is disrupting software production at a scale we've never seen. On the other, it's commoditizing many of our established ways of building apps—destabilizing the whole value chain and temporarily offsetting the theoretical productivity gains everyone expected, as GitLab's survey suggests.
And yet, I believe three mistakes are right in front of us.
Dismissing what's happening just because the productivity gain isn't guaranteed. Lack of immediate ROI doesn't invalidate a paradigm shift—it rather often signals it.
Imposing optimization logic too early, suffocating the necessary adjustment of the "production system"—often through budget cuts—at the very moment when workflows need to be redesigned, not squeezed.
Falling into a new fatalism, assuming that only top-tier engineers will be able to master AI-assisted development. That's the fastest path to creating a new divide instead of enabling everyone to climb the learning curve.
In a way, this moment looks a lot like the early days of the automobile. Early motorists of the 1900s were pioneers who could do everything: fix the engine on the roadside, tune the carburetor, improvise repairs. None of this meant the automobile wasn't destined to scale. In fact, the car wasn't a "better horse"—it was a completely different paradigm. And like any paradigm shift, its early phase was chaotic, inefficient, and reserved to those who could navigate the complexity before tooling and standardization made it accessible to everyone.
AI-driven software engineering is possibly hitting a similar curve today—a messy, pioneering phase that precedes a real, widespread productivity shift. The teams that focus on shaping the environment, not just stacking tools, will be the ones who unlock it.
