AI
AI Hasn't Made Engineers More Productive—Just More Drained
While AI coding tools promise speed and scale, engineering leaders warn that shipping more code faster doesn't equal better business outcomes, and the constant pressure to manage AI agents is burning out teams.
The New Stack
Industry leaders increasingly argue that AI has fundamentally shifted what it means to be a software engineer. Rather than writing code in isolation, engineers now resemble front-line managers juggling competing priorities, coordinating across teams, and shepherding AI agents through their work—a view echoed by recent research predicting organizations will move toward "tiny teams" of just two to three people by 2029.
Yet despite this explosion in tooling and the promise of 10x productivity gains, technologists are grappling with an uncomfortable paradox: engineers report feeling simultaneously more productive and more exhausted.
Output ≠ Outcome#
The disconnect lies partly in what companies are actually measuring. Most organizations still track the same metrics they used before AI arrived—lines of code, pull requests merged, and velocity points—even though these signals fail to capture whether the extra output delivers real value.
As Ameya Kanitkar, founder of AI measurement platform Larridin, explains, rewarding code volume and commit frequency can actually push teams away from meaningful customer impact. When AI can inflate these metrics effortlessly, engineers may chase volume at the expense of solving real problems.
Daniel Wang, CTO at Citizen Health and a former engineering director at Uber, frames the issue plainly: "Software exists to solve problems. If AI lets us ship 10x as much code but customer outcomes don't improve, we aren't more productive." Better metrics would track cycle time from concept to production, rollback rates, defect escape, system reliability—and crucially, whether the final product actually solves customer problems.
The Hidden Cost of Agent Management#
The fatigue engineers feel stems partly from a new responsibility: constant agent babysitting. As Kanitkar notes, teams now context-switch rapidly to feed work to multiple AI agents, managing their output and keeping them productive. Some engineers even run agents overnight to have finished work waiting in the morning—a practice that sounds efficient but creates relentless pressure to maintain a pipeline of tasks.
David Holz, founder of Midjourney, recently observed this tension on X: engineers feel "extremely productive and also extremely drained" with newer coding models, signaling that something in the equation has gone wrong.
