Cognition CEO Scott Wu: AI Coding Agents Should Augment, Not Replace Developers

In a year when nearly every major tech CEO has announced layoffs tied to AI automation, Cognition's Scott Wu is delivering a sharply different message: AI coding agents are here to help you build more, not take your job.

"We've never thought about it as replacing humans," Wu said in an interview with.¹ "I know it's like a scenario, folks have said these things. It has never been our view."

The comments come just two days after his two‑year‑old startup raised more than $1 billion at a $26 billion post‑money valuation — more than doubling its $10.2 billion valuation from just eight months earlier. Cognition's annualized revenue run rate has hit $492 million, with enterprise usage of its flagship AI coding agent Devin growing 50% month‑over‑month for six straight months.

Wu describes Devin's current skill level as "somewhere between a junior and a mid‑level engineer," depending on the task. The agent excels at what developers often find tedious: long‑tail maintenance, updating legacy software, and platform migrations — the kind of "toil" work that burns out engineering teams.

"When we started building Devin, it's kind of a funny thing ... we really just thought of it as: this is your buddy who helps you build more," Wu told,¹ noting he keeps a stuffed‑animal Devin on his desk.

The internal numbers at Cognition tell a striking story: 89% of code committed by the company's own engineers is authored by Devin, with the remainder coming from Windsurf, a competitor Cognition.² Yet Wu frames this as a partnership, not an omen.

Cognition's enterprise customer list reads like a who's‑who of large organizations adopting AI coding tools.,² clients now include Mercedes‑Benz, Goldman Sachs, Citi, Santander, NASA, Dell, the U.S. Army, and the U.S. Navy.

The results are dramatic. Mercedes‑Benz used Devin to cut an 8‑month legacy modernization project down to just 8 days. Latin America's largest bank, Itaú, reports that Devin automatically fixes 70% of security vulnerabilities. System integrators Infosys and Cognizant have embedded Devin directly into their delivery workflows.

"Cloud agents have gone from niche to mainstream, and today they are the fastest growing way to create software," Cognition wrote in its.²

Wu's defensive posture toward developers isn't surprising given his background. A profile in ¹ described him as "one of the most accomplished child competitive programmers of all time." He won a nationwide math competition for seventh‑graders as a second‑grader and started coding at age nine.

"We are all programmers ourselves. I started coding when I was nine," Wu told.¹ Through childhood coding tournaments he met other future AI founders, including Alexandr Wang of Scale AI.

"It's not a secret, most software engineers love building software, right? ... I can make my whole idea that I have, and turn it into a product. I can turn it into an experience," Wu said, explaining why preserving the creative joy of programming matters to him.

Cognition positions itself as model‑agnostic — an "agent lab" that works with all foundation model providers rather than being locked into one. The company evaluates models across more than 100 categories of software engineering tasks and recently launched its own model training program.

Its first proprietary model, SWE‑1.6, is already the most‑used model inside Windsurf, Cognition's IDE with integrated agents. Customers praise its speed — up to 950 tokens per second — and cost efficiency. According to Cognition, Devin's architecture automatically balances price‑to‑performance by routing different tasks to different models.

"The value of independence is increasing as token usage grows exponentially across the industry," the company stated.

Wu expects coding agents to be the first wave of agentic AI, with customer service, medicine, and other fields following. He describes the trend toward self‑improving software — AI that writes better AI — as a "wild ride" ahead.

Yet he's insistent that human oversight remains paramount across all industries. The human, Wu argues, should always decide what to do — the agent handles execution. He sees AI as another layer of abstraction in the long arc of software engineering, similar to how visual development environments once abstracted away machine instructions.

As TechCrunch notes, this philosophy cuts against the grain of an industry where daily layoff announcements cite AI as the reason. Whether that message holds as Devin's capabilities grow from junior to senior — and beyond — is the question builders will be watching.