Cognition’s Scott Wu asserts that AI coding agents ought not to substitute for humans

Cognition’s Scott Wu asserts that AI coding agents ought not to substitute for humans

Cognition CEO Scott Wu has made waves once more this week as his AI coding agent startup, just two years old, secured $1 billion in funding at a valuation of $26 billion. Cognition is known for Devin, among the pioneering and arguably most effective AI coding agents available. According to the CEO, Devin “naturally manages tasks from start to finish.”

In the recent blog post announcing the funding, Cognition outlined a vision of a future where “we are transitioning to a realm of self-driving software development.”

So, is it possible for Devin to take over the role of a mid-level L4 programmer? Yes and no, Wu stated to TechCrunch. “We’ve never seen it as a replacement for humans. I understand that this is a common narrative, but it has never been our perspective.”

In this chaotic year of 2026, where it seems like every day another tech CEO declares job cuts in the name of replacing employees with AI, Wu emphasizes his desire for coders to retain their jobs. “We are all programmers ourselves,” he remarked. “I started coding at the age of nine.”

Indeed, Wu has been recognized as one of the most skilled child competitive programmers in history, as highlighted in a recent piece by Colossus. In second grade, Wu claimed victory in a nationwide math contest for seventh-graders, which led to a youth filled with math and programming competitions. This also connected him with other prodigies who would later establish AI tech startups, including Scale AI founder Alexandr Wang.

Thus, he explains to TechCrunch, the intention has never been to render human programmers redundant.

“When we initiated the development of Devin, it’s a bit amusing,” he reflected, “but we genuinely regarded it as: this is your companion who assists you in building more.” He even showcased a small stuffed animal holding a computer, his own version of a Devin teddy bear, which he keeps on his desk. He considers it a tangible representation of the Devin AI coder “This is my friend that helps you create more.”

Wu does not wish for AI agents to diminish the joy of programming for people.

“It’s no secret that most software engineers enjoy crafting software, right?” he noted. “If you ask them why, they’ll generally respond, ‘Well, it’s like I get to create things from nothing. I can take my entire concept and transform it into a product. I can convert it into an experience.’”

Similarly to how visual development environments abstracted software creation from machine instructions, he perceives agents as another level of abstraction between imagining a software product and realizing it.

Nevertheless, Cognition claims that Devin’s primary function within the company is to produce nearly all software. The firm asserts that 89% of code committed by its engineers was accomplished by Devin, with the remainder attributed to local agents in Windsurf, the AI coding challenger they acquired last year.

Wu elaborates that his agent primarily handles the long-tail maintenance tasks that many programmers typically dislike: updating outdated software; migrating applications from one platform to another. Agents will liberate programmers “from much of the drudgery, enabling them to engage much more in creative work,” he assures.

Thus, Wu is resistant to the notion of Devin “replacing” human programmers. He acknowledges that while it can function independently, its performance aligns “somewhere between a junior and a mid-level engineer” based on the task requirements.

Regarding the idea of self-driving software, where the agent learns and enhances itself, potentially enabling it to operate at higher levels (“recursive” being the trending term in AI these days), Wu states, “I believe we are in for an exhilarating journey.”

He envisions agents making their way into additional sectors where they’ll acquire tasks, from customer support to healthcare, but hopes that the objective remains to augment human workers in those areas as well.

“Coding and software have been the first to evolve, but we will observe this phenomenon in all these other fields,” he forecasts. “One thing that has been clear to us since the beginning is, it should always be up to the human to decide what to do … you can really see this in software engineering, but I think it applies to all these other professions as well.”

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.