Cobus Greyling, Chief Evangelist at Kore.ai, identifies a recurring pattern of developer stack layers collapsing under the effect of AI agents. After the collapse of the integration layer (the CLI replacing MCP as the natural interface for agents) and the orchestration layer (natural-language prompts in a CLAUDE.md file replacing frameworks such as LangGraph or CrewAI), it is now the development environment itself that is collapsing: the IDE becomes optional.
Greyling breaks down the five essential functions of an IDE - file navigation, code editing, syntax awareness, build/run/debug, version control - and demonstrates that each is perfectly achievable by an agent via the CLI, without a visual interface. The agent navigates by intent (grep, find, glob) rather than by clicking through a file tree. It parses raw code natively thanks to its training on billions of lines of code, without needing syntax highlighting. It reads stdout/stderr to debug in a single loop, without breakpoints. It understands diffs in plain text without a graphical viewer.
The central argument is an inversion: the IDE was built to make the command line accessible to humans, compensating for their cognitive limitations (difficulty parsing raw text, need for spatial navigation, inability to hold an entire call stack in working memory). An AI agent has none of these limitations. The context window IS its working memory. Raw text IS its native interface.
Greyling nuances his argument by identifying four cases where the IDE remains relevant: when visual output matters (UI design), for architectural refactoring requiring human judgment across many files, for onboarding onto a new codebase, and for compliance/audit workflows. But for day-to-day development work - writing features, fixing bugs, running tests, committing code - the IDE represents unnecessary overhead for the agent.
The consequence is a fundamental shift in the developer's role: competence shifts from mastery of the tool (keyboard shortcuts, extension ecosystem, debugger workflows) to the ability to clearly describe intent, critically review outputs, and understand the architecture well enough to guide the agent. The interface between the human and the code shifts from visual manipulation to verbal specification.