The modern software development cycle has long been defined by a rigid wall between the people who decide what to build and the people who actually write the code. For years, the handoff from product management to engineering was a bottleneck of documentation, misinterpreted requirements, and endless ticket grooming. But a shift is occurring in how teams interact with their project management tools, moving away from static tracking and toward an era of autonomous execution. The friction that once existed between a product requirement and a pull request is beginning to evaporate.

The Quantifiable Explosion of AI Output

Data from Linear reveals a massive acceleration in AI adoption across every organizational layer between January and June 2026. The growth is not limited to the engineering department; it is permeating the entire company structure. Active AI users in product management roles jumped from 12% to 34%, while go-to-market teams saw their usage climb from 5% to 18%. Perhaps most telling is the behavior of executive leadership. CEOs of companies with more than 201 employees saw their AI usage surge from 9% to 36%, suggesting that high-level leadership is moving away from reading summarized reports and toward direct interaction with AI tools to understand their product's trajectory.

This adoption has triggered a vertical spike in actual output. When comparing current data to June 2024, the number of pull requests generated per workspace has increased by 111%. The disparity becomes even more stark when isolating the impact of coding agents. Teams that integrated AI coding agents saw their weekly pull request volume skyrocket from 21 to 65. In contrast, teams operating without agents saw only a marginal increase, moving from 8 to 10 pull requests per week. The gap is no longer about individual developer speed, but about the systemic capacity provided by agentic workflows.

This shift extends to the very inception of work. Two years ago, AI-generated issues were a rarity, accounting for fewer than one in every 1,000 tickets. Today, nearly half of all issues created within Linear are authored by AI. At the current trajectory, AI is poised to generate more work items than humans and integrated tools combined, fundamentally changing how backlogs are populated and managed.

The Jevons Paradox and the Rise of the Builder

Conventional wisdom suggests that AI efficiency should lead to a reduction in total working hours. However, the reality within Linear's ecosystem suggests the opposite. While the time spent on high-level decision-making—such as processing customer requests, drafting documentation, and project planning—has remained stagnant over the past year, the time spent on execution has increased. Specifically, time spent on issue creation, classification, and commenting has risen across almost all roles, with engineering teams seeing a 17% increase in time dedicated to these tasks.

This phenomenon is a classic manifestation of the Jevons Paradox: as a resource becomes more efficient to use, the total consumption of that resource actually increases rather than decreases. In the context of software development, AI has not given developers more free time; it has lowered the cost of execution so significantly that teams are simply doing more. The total time invested in product development has increased because the barrier to initiating and refining a feature has collapsed.

This collapse is erasing the traditional boundaries of professional roles. The data shows a growing number of non-engineers stepping directly into the codebase. The percentage of product managers who directly attach pull requests to their work has risen from 3% to 10%, while designers have seen a similar jump from 1% to 8%. These individuals are no longer just describing changes in a ticket; they are using AI to commit code and deploy changes. The organization is shifting from a structure of specialists to a culture of builders, where the ability to execute is no longer gated by syntax proficiency.

This evolution necessitates a complete overhaul of how performance is measured. For too long, the industry has focused on token consumption or the number of AI prompts as a proxy for productivity. But token volume is a vanity metric; it can be inflated by low-value tasks like mechanical refactoring. The new gold standard for AI contribution is actual output—specifically, the volume and quality of pull requests and the velocity of issue resolution.

As the role of the engineer shifts from primary implementer to architect and validator, the focus moves toward managing the expanded agency of the non-technical builder. The challenge is no longer about saving time, but about managing the massive increase in output and ensuring that the surge in pull requests translates into a more polished product rather than just more code.