AI is a standard tool for software developers, but business results are not improving at the same pace. Laura Tacho’s presentation at the Pragmatic Summit put numbers on it. About 93% of developers use AI to help write code, and roughly a quarter of the code that ships in live products is written by AI. Yet in data covering 121,000 developers, measured productivity gains are around 10%. With usage that high and gains that small, the constraint lies in how work is defined and moved through teams.
Professional developers work inside established systems of approvals, reviews and handoffs between people and teams. AI can speed up one step and leave the total time to deliver mostly unchanged if the rest of the process stays the same.
Non-technical builders show how much rests on defining the work. Cynthia Chen built an iOS app that identifies dog breeds. Laura Zaccaria and Wei Khjan Chan built a family meal planner and a tool for filing receipts and expenses. None of them relied on deep technical expertise. Each described what needed to happen in plain language, then broke the work into manageable steps. They were specific, kept refining and rephrased the task when the AI got stuck.
Many teams have instead added AI to existing habits without changing how work is set up at the start. When the task is unclear, AI produces more noise faster. When the task is clearly defined, AI can shorten the path from problem to working result.
Leaders invest in AI for speed, lower cost and more delivery capacity, so counting licences, usage or generated code does not show who is getting value. A better measure is whether a team can pick a painful problem, define the outcome clearly and produce something useful in a short cycle.
The advantage in 2026 belongs to people who can explain a task as if to a smart new hire on day one. They say what the situation is, what “done” looks like and what to avoid. That speeds work up, because less of it is redone after fixes. It also improves quality, because fewer assumptions are left unstated.
