跳到正文
Business Insider Tech·· 3 小時前AI 評分51

Cognition 聯合創辦人稱 AI 編碼已達「智能飽和」

AI coding has hit an 'intelligence saturation' point, Cognition cofounder says

AI 導讀

Cognition 聯合創辦人兼總裁 Russell Kaplan 表示,愈來愈多軟件工程任務已達「智能飽和」,模型能做好任務後,成本和速度便更重要。Kaplan 説,這與 Cognition 內部數據的趨勢一致;The Information 報道,模型智能領先 10 點帶來的開發者流量增幅,已由 2025 年末近 69% 降至 20%。

正文

By Ben Shimkus

You're currently following this author! Click to unsubscribe from email alerts.

Digital advertisement for Cognition's Devin AI software engineer beside escalators near Exit B2 on the concourse at Embarcadero BART and Muni Metro station, San Francisco, California, August 27, 2026.

The cofounder of Cognition, the company behind Devin, says the AI model competition is now favoring speed over intelligence. Smith Collection/Gado/Getty Images

For years, the AI race has favored the smarter model.

Russell Kaplan, the president and cofounder of AI coding startup Cognition, the company behind software-engineering platform Devin, said that could be changing for a growing number of software-engineering tasks.

"Many tasks within software engineering are now 'intelligence saturated,'" Kaplan wrote in a post on Monday. "Once any model can do the task well, cost and speed are all that matter."

This is directionally consistent with our internal data @cognition. Many tasks within software engineering are now "intelligence saturated." Once any model can do the task well, cost and speed are all that matter.

— Russell Kaplan (@russelljkaplan) October 19, 2026

Kaplan was responding to an article from The Information. Their reporting found that a 10-point advantage in model intelligence now delivers just 20% more developer traffic, down from nearly 69% in late 2025.

He said that story is "directionally consistent" with what is happening inside Cognition.

That could be a sizable strategic shift for the AI companies fighting to win software developers and enterprise contracts. OpenAI, Anthropic, Meta, Google, and a growing crop of AI labs have all highlighted their increasingly capable coding tools as a reason to choose their models.

Now, as those AI agents get better at routine tasks, one model's slightly smarter approach could matter less to users if another model is cheaper and faster.

Kaplan didn't immediately respond to a request for comment from Business Insider.

Read next

Business Insider tells the innovative stories you want to know

Business Insider tells the innovative stories you want to know

Business Insider tells the innovative stories you want to know

Business Insider tells the innovative stories you want to know

Ben Shimkus

You're currently following this author! Click to unsubscribe from email alerts.

Ben Shimkus is a reporter for the Business News desk. He writes about cars, transportation, retail, and jobs. Ben's reporting has appeared in Rolling Stone, The Verge, Automotive News, USA Today, AutoBody News, LGBTQ Nation, TopSpeed, and Out Magazine. He's also held staff writing positions at The U.S. Sun and the Daily Mail. He graduated from NYU with a Master's in journalism in 2024. Email Ben at bshimkus@insider.com or message him privately on Signal at bshimkus.41. 

來源:Business Insider Tech · businessinsider.com