Linear CEO Karri Saarinen has published a detailed article discussing the real-world progress of AI tools, stating that agentic coding—a method where AI Agents participate in planning, modifying code, debugging, and submitting changes—is now quite common.
Data from Linear reflects this shift: a majority of their pAId workspaces have installed code Agents, and related usage ACTivity has grown more than fivefold in just a few months. Linear's own cloud-based code agent now resolves over 1,000 issues per month, a number that continues to grow rapidly.
Saarinen offers a measured assessment of this trend. He notes that very few people privately claim agents can write 100% of the code, and it's rare to hear of real companies running fully autonomous agent clusters at scale. Engineers still need to provide direction, constraints, and judgment, typically manAGIng only a few local agents while offloading minor fixes to a small number of cloud agents running in the background.
He summarizes the primary impact of AI on programming as "increasing bandwidth": tasks that were previously too small, too tedious, or too time-consuming are now much easier to handle. However, genuinely difficult problems are not being accelerated to the Same degree; they still require a deep underStanding of the system, trade-off analysis, and judgment about what should exist.
Saarinen also believes that AI will not render planning obsolete. As the cost of building decreases, the risk of building the wrong thing becomes greater. Teams still need to decide what is most important, although planning cycles may need to be shorter and allow more room for experimentation.
Comments & Questions (0)
No comments yet
Be the first to comment!