AI coding is becoming a stack
Industry coverage argues that AI coding tools are not settling on a single assistant but are instead forming a layered “AI coding stack” with separate orchestration, execution and review roles. The New Stack lays out how Cursor, Claude Code and Codex are converging into composable stages, while The Verge frames the contest as firms fighting to own the broader development workflow rather than just models. (thenewstack.io) (theverge.com)
AI coding tools are settling into layers, not one winner: one product plans work, another runs it, and a third checks it. (thenewstack.io) The New Stack reported on April 12 that Cursor, Anthropic’s Claude Code, and OpenAI’s Codex are starting to work like a stack with separate orchestration, execution, and review roles. It pointed to one week in early April when Cursor shipped a rebuilt multi-agent interface and OpenAI published a Codex plugin for Claude Code. (thenewstack.io) Cursor’s part of that stack looks like a control panel. Cursor 3, released April 2, replaced its Composer pane with an Agents Window for running parallel agents across local machines, worktrees, and cloud sandboxes, while Cursor’s pricing page now sells cloud agents and a separate Bugbot review product. (thenewstack.io) (cursor.com) Claude Code’s role is closer to the hands on the keyboard. Anthropic says the tool can read a codebase, edit files across a project, run tests, use command-line tools, and work from the terminal, desktop app, integrated development environments, Slack, and the web. (anthropic.com) (claude.com) Codex is being pitched as both a worker and a reviewer. OpenAI says Codex handles feature work, refactors, migrations, reviews, and releases, and its app is designed for multi-agent workflows with built-in worktrees and cloud environments. (openai.com) The competition has widened from model quality to workflow ownership. OpenAI said on April 2 that teams on ChatGPT Business and Enterprise can now add Codex-only seats with pay-as-you-go pricing, and that Codex users inside those workspaces have grown 6 times since January while more than 2 million builders use Codex every week. (openai.com) Anthropic is making a similar push into broader software work. Its Claude Code product page says the majority of Anthropic’s own code is now written by Claude Code, and its commercial plans package the tool into Pro, Max, Team, and Enterprise tiers instead of treating it as a narrow add-on. (anthropic.com) (claude.com) Cursor is also moving beyond an editor into a platform that mixes model access, background agents, governance controls, and code review. Its current plans range from free to $200 a month for individuals and $40 per user a month for Teams, with SAML single sign-on, usage analytics, and privacy controls aimed at company rollouts. (cursor.com) The boundaries between rivals are getting looser inside the tools themselves. The New Stack reported that OpenAI’s codex-plugin-cc installs inside Claude Code and can review Claude’s output or hand a task to Codex as a subagent, while JetBrains added native Codex access to its integrated development environments in January. (thenewstack.io) (blog.jetbrains.com) That leaves developers choosing combinations instead of allegiances. The shape emerging in April 2026 is less one chatbot replacing the editor than a software toolchain where planning, execution, and review can come from different companies in the same session. (thenewstack.io) (openai.com)
