The fierce competition driving the artificial intelligence industry to develop ever-smarter models may be reaching a turning point in the software-engineering sector. According to Russell Kaplan, president and cofounder of AI coding startup Cognition, a growing number of software-engineering tasks have reached a state of “intelligence saturation.”
Cognition is the company behind Devin, a software-engineering platform designed to automate complex coding workflows. Writing in a public post, Kaplan stated that once an AI model achieves the capability to execute a specific task well, raw intelligence ceases to be the primary differentiator for users. Instead, factors such as cost and operating speed take precedence.
Kaplan’s remarks were made in response to published reporting indicating that a 10-point advantage in model intelligence now yields only about 20 percent more developer traffic. That figure represents a substantial drop from late 2025, when a similar intelligence margin drove nearly 69 percent more traffic. Kaplan noted that this downward trend aligns directly with internal data gathered at Cognition.
This shift could signal a major strategic recalculation for major technology firms and specialized artificial intelligence laboratories. Companies including OpenAI, Anthropic, Meta, and Google have heavily marketed their increasingly sophisticated coding agents, routinely pointing to incremental gains in model intelligence as the primary justification for enterprise clients and developers to adopt their ecosystems.
As foundational models continue to improve at handling routine and repetitive coding tasks, the strategic landscape is evolving. When multiple competing models are capable of performing a given task with comparable competence, minor advantages in reasoning capabilities may matter far less to buyers if a competing model offers faster execution speeds or a lower price point. Kaplan did not immediately respond to requests for further comment.
