OpenAI turns Codex import into a live sync aimed at Claude Code shops

OpenAI has quietly rebuilt the walls around enterprise coding agents, in the other direction. Over the past week the company expanded its import tooling so that Codex and the ChatGPT desktop app can pull projects, chats, skills, plugins, and configuration from competing agents, then keep the imported material synchronized with the original.

OpenAI's import documentation spells out the scope. The desktop app imports from Claude Code, Claude Cowork, or Cursor. Codex CLI imports from Claude Code or Cursor via a /import command. What comes across is close to a full working environment: AGENTS.md instruction files, config.toml settings, skills, plugins, MCP server configurations, project folders and project memories, hooks, slash commands, subagent definitions, and chat sessions from the previous 30 days, capped at 50 chats on the CLI. The docs stress the process leaves the source agent untouched.

Sync is the new part

One-time migration tools are old news. The change that matters is the sync toggle: desktop users can, in OpenAI's words, turn on automatic updates to keep imported work in sync with the original agent. The Codex changelog shows the plumbing arrived in CLI 0.147.0 on August 7, which added synchronization of changes to imported Claude and Cursor conversations without creating duplicates, building on July's expansion of the /import command. In practice a developer can now run Codex alongside Claude Code for weeks, with Codex continuously absorbing the work happening next door. Migration stops being an event and becomes a gradient.

Demand for this was visible in public. A feature request open since May asked for exactly this kind of project and chat migration, arguing that without it, moving to Codex meant manually reconstructing months of organized work.

Switching costs are the moat

For GaaS buyers this is the most consequential kind of feature: one that changes vendor leverage rather than model capability. Enterprise agent deals are sticky because the accumulated context, the instruction files, the memories, the tool configurations, is expensive to recreate. OpenAI is systematically zeroing out that cost for inbound switchers at exactly the moment it is pushing ChatGPT Work into large accounts, a campaign we covered when Codex and ChatGPT Work crossed 10 million business seats. A procurement team running a bake-off can now demand a live side-by-side with real project history in both tools, then compare consumption against the controls we detailed in our report on Codex token budgets.

The door swings both ways, and Anthropic has not stood still; release trackers show Claude Code shipping its own import of Cursor-managed skills and synced conversations in recent builds. That symmetry is the real story. The coding agent market has concluded that lock-in through data gravity is no longer defensible, so the competition moves to whichever agent does better work on the same imported context. For enterprises negotiating renewals this quarter, the practical takeaway is simple: the cost of leaving your incumbent coding agent just dropped to roughly zero, and your vendor knows it.