Cognition’s Scott Wu argues AI coding agents should assist developers, not replace them

Cognition co-founder Scott Wu is pushing back against the idea that AI coding agents should replace human programmers, even as his company’s Devin system is one of the best-known examples of the technology. In comments highlighted by TechCrunch, Wu said the agent is intended to assist developers rather than supplant them, underscoring a growing debate over how far AI should go in software engineering.

The distinction matters because AI coding tools are moving quickly from novelty to everyday workplace software. TechCrunch reported that developers are increasingly relying on AI to write code faster, but researchers warn that speed does not necessarily translate into better code, raising concerns about quality, maintainability and long-term risk. That tension sits at the center of Wu’s argument: AI can automate parts of coding, but it should not be treated as a full substitute for human judgment.

The conversation is part of a broader shift in how the tech industry talks about artificial intelligence. Bloomberg reported that corporate language has moved from “generative” AI to “agentic” AI, reflecting a new emphasis on systems that can take actions and complete tasks with less supervision. IBM and Thomson Reuters describe agentic AI as software designed to make decisions, plan steps and act on goals, while generative AI is mainly focused on creating content in response to prompts.

That shift has also fueled a more aggressive view inside some companies that AI can replace workers outright. TechCrunch reported in a separate piece that some leaders are becoming “too AI-pilled,” with Box founder Aaron Levie warning that the people deciding AI can replace jobs are often the least familiar with what those jobs actually involve. The same report cited ClickUp cutting 22% of its workforce for AI agents, a sign that automation is already affecting headcount decisions.

Against that backdrop, Wu’s remarks reflect a more cautious approach from one of the most prominent names in AI coding. Rather than arguing that agents can fully do a programmer’s job, the message from Cognition is that they are tools for productivity and support. That framing aligns with the view from IBM, Thomson Reuters and Red Hat that agentic systems are meant to augment human work, not eliminate the need for human oversight.

The issue is likely to remain contentious as companies continue experimenting with AI in software development and other white-collar fields. Bloomberg said markets are still betting heavily on AI growth, while researchers and industry voices are increasingly warning about overconfidence in the technology’s ability to replace people. For developers, the immediate question is not whether AI can write code, but how much trust teams should place in it — and where human engineers still need to stay firmly in the loop.