Markets
AI Coding Tools Are Hitting a Ceiling Where Speed and Price Beat Raw Intelligence
By Bear Creek Staff, . Bear Creek Capital Network.
Russell Kaplan, president and cofounder of AI coding startup Cognition, said this week that the race to build smarter coding models may already be losing its edge for a growing share of software-engineering work.
"Many tasks within software engineering are now 'intelligence saturated,'" Kaplan wrote in a post on Monday, according to Business Insider Markets. "Once any model can do the task well, cost and speed are all that matter."
Kaplan was responding to reporting from The Information, which found that a 10-point lead in model intelligence now produces roughly 20% more developer traffic, down from close to 69% in late 2025. He described that finding as "directionally consistent" with what Cognition is seeing internally.
The practical meaning: when every major model clears a competence threshold on a given task, the differentiator stops being capability and becomes price and latency. A model that is marginally smarter loses to one that is meaningfully cheaper or faster.
What This Means for the Big Labs
OpenAI, Anthropic, Meta, Google and a widening field of AI labs have all leaned on coding performance as a primary reason for developers and enterprise buyers to choose their products. That pitch works as long as intelligence gaps between models translate into real productivity differences for users.
If Kaplan's framing is right, that translation is weakening, at least for routine tasks. The competitive pressure shifts toward infrastructure efficiency and pricing strategy rather than benchmark scores. Labs that have invested heavily in frontier model capability may find that advantage harder to monetize on the coding side than their roadmaps assumed.
What the Reporting Does Not Establish
Several things remain open. Kaplan's comments describe Cognition's own internal data but do not specify which categories of software-engineering tasks have reached saturation, or what share of total developer work they represent. The distinction matters: complex, multi-step engineering problems may still reward raw intelligence, while repetitive or well-defined tasks may not. No independent verification of The Information's traffic figures was included in the Business Insider report.
Kaplan did not respond to a request for comment from Business Insider before publication.
The harder question for investors and enterprise buyers is whether saturation spreads upward into more complex tasks over time, or whether it stays concentrated at the routine end. If it spreads, the pricing and speed advantages held by leaner, faster models become structurally more important. If it stays narrow, frontier capability still commands a premium where it counts. That answer is not in the current reporting, and it is probably the one worth watching.
Source: https://www.businessinsider.com/cognition-cofounder-ai-race-intelligence-saturation-2026-10
Bear Creek Capital Network publishes content for informational and educational purposes only. Nothing in this article constitutes financial, investment, tax or legal advice, or a recommendation to buy or sell any security. Consider your own circumstances and consult a qualified professional before making financial decisions.