Linear CEO: AI Code Agents Increase Bandwidth, Not Replace Engineers

AIMPACT Update, April 27 (UTC+8): According to monitoring by Beating, Linear CEO Karri Saarinen published a detailed post on the real progress of AI tools, stating that agentic coding—where AI agents participate in planning, modifying, debugging, and submitting code—is now widespread. Linear’s own data reflects this shift: the majority of paid workspaces have installed code agents, with related usage activity growing more than fivefold in just a few months. Alone, Linear’s cloud-based code agents are now resolving over 1,000 issues per month, and this number continues to grow rapidly. Saarinen’s assessment of this transformation is measured. He notes that privately, almost no one claims agents can write 100% of code, and few real companies are running fully autonomous agent clusters at scale. Engineers still provide direction, constraints, and judgment; typically, they manage only a few local agents directly, while delegating a small number of cloud agents to handle minor fixes in the background. He summarizes AI’s primary impact on programming as “increasing bandwidth”: problems that were previously too small, too annoying, or too time-consuming are now easier to address on the fly. However, genuinely difficult problems have not been accelerated to the same degree—they still require deep system understanding, trade-off analysis, and judgment about what should exist. Saarinen also believes AI will not diminish the value of planning. As the cost of building decreases, it becomes easier to build the wrong things; teams still need to determine what matters most, but planning cycles must shorten and more room must be made for experimentation. (Source: BlockBeats)