Claude Code creator says agentic coding is changing the developer’s job

A widely shared Chinese-language report has put a sharper spotlight on a question many software teams are already asking: if coding agents can now work across entire repositories, what exactly does a developer do next?

The report, published by the WeChat account 图灵智语, centers on remarks attributed to Boris Cherny, creator of Claude Code, during a Y Combinator podcast appearance. According to the article, an X user summarized Cherny as saying that since Opus 4.5 he had removed his IDE from his daily workflow and no longer edits individual lines of code by hand, with Claude producing the code instead.

Claude Code creator says agentic coding is changing the developer’s job

From editor to workflow cockpit

The claim is striking because it comes from someone closely associated with the tool itself. But the larger point is not that programming knowledge has become irrelevant. The article argues that the developer’s interface is moving away from typing every line in a traditional editor and toward designing the loop in which an AI system plans, edits, tests, receives feedback, and tries again.

That framing matches how Anthropic presents Claude Code in its own documentation. The product is described less as a simple autocomplete assistant and more as an agentic coding environment that can understand a codebase, operate across files, use tools, resume context, run in parallel worktrees, and support batch or CI-style workflows.

What changes for engineers

In this model, the scarce skill is not merely syntax recall or typing speed. It is the ability to define a task boundary, provide the right repository context, set acceptance criteria, create tests, review generated changes, and decide when the system’s output is safe to ship. Developers become closer to workflow architects, reviewers, and operators of automated engineering systems.

The report also highlights “loops” and “dynamic workflow” as key concepts. A mature coding-agent process is not a one-shot prompt. It is a controlled cycle: plan first, make a bounded change, run checks, inspect failures, revise, and keep a human responsible for the final judgment. Parallel sessions and subagents can speed up research or implementation, but they also increase the importance of version control, isolation, and rollback discipline.

Why the discussion matters

The debate arrives as AI coding tools are being repositioned from productivity add-ons into full development surfaces. If that shift continues, IDEs may remain important, but the center of gravity for many tasks could move to terminal agents, browser-based agents, CI pipelines, and project-specific automation scripts.

For ordinary developers, the takeaway is practical rather than apocalyptic. Coding fundamentals still matter because someone must understand, test, and own the result. But the daily work may increasingly reward people who can translate product intent into precise tasks, manage context efficiently, and build reliable quality gates around AI-generated code.