Boris Cherny, the creator and head of Claude Code at Anthropic, published a thread on X in early January 2026 outlining his personal development workflow, prompting sustained discussion across the engineering community. The post, which described running five instances of Claude simultaneously in a terminal, arrived as Anthropic separately disclosed that Claude Code had reached $1 billion in annual recurring revenue.

The timing matters. Anthropic is operating in a market where OpenAI, Google, and a growing field of agentic coding tools are competing for developer mindshare. Claude Code, which launched as a terminal-native coding agent, has positioned itself as a tool for professional engineers rather than casual users. Cherny's disclosure offered a rare, first-hand account of how the product's own creator uses it in practice, lending the workflow a credibility that marketing materials cannot replicate.

The workflow Cherny described centres on parallelism. He runs five Claude agents concurrently across numbered terminal tabs, using iTerm2 system notifications to manage handoffs between sessions. He also runs a further five to ten instances via claude.ai in a browser, using a "teleport" command to move sessions between the web interface and his local machine. Notably, he uses Anthropic's heaviest model, Opus 4.5, for all tasks, arguing that its superior tool use and reduced need for correction makes it faster in practice than lighter alternatives despite higher latency per token. His team also maintains a shared file called CLAUDE.md in their git repository, which accumulates instructions each time the agent makes an error, effectively encoding institutional knowledge into the agent's context over time. A custom slash command, /commit-push-pr, handles version control bureaucracy autonomously, and dedicated subagents manage code simplification and end-to-end verification before anything ships.

The reaction to the thread points to a broader shift in how professional developers are beginning to think about AI tooling. The workflow Cherny describes is not an autocomplete function bolted onto an editor; it is closer to a task-dispatch system in which a single engineer coordinates multiple autonomous work streams. That framing aligns with comments made by Anthropic President Daniela Amodei earlier in January, in which she described a "do more with less" thesis for the company's product direction.

Whether the workflow is reproducible at scale across engineering teams of varying sizes and disciplines remains an open question. The CLAUDE.md approach, for instance, depends on consistent human review of agent output, which introduces its own overhead.