Boris Cherny, the engineer who built Claude Code, says that on some days he is actively managing tens of thousands of AI agents running simultaneously. The disclosure, made in an interview with Fortune, illustrates just how far agentic AI workflows have already moved beyond laboratory experiments and into daily professional practice at Anthropic itself.
A Glimpse Into Agentic Scale
Cherny's comments are striking because they come from someone who uses Claude Code as a practitioner, not just its designer. He has previously described going months without writing code directly, relying instead on AI agents to handle the work. Now, the scale he is describing suggests that multi-agent orchestration has become a core part of how some engineers at Anthropic operate. For context on how Cherny's own relationship with coding has shifted, Claude Code's Creator Hasn't Written a Line of Code in Eight Months gives a fuller picture of that transition.
Key Facts
- Boris Cherny created Claude Code, Anthropic's agentic coding tool.
- He reports managing tens of thousands of AI agents simultaneously on some days.
- The scale of orchestration points to rapid maturation of multi-agent workflows.
- Cherny has described not writing code directly for an extended period, delegating tasks to agents instead.
- The comments were made in an interview published by Fortune.
The numbers Cherny cites are significant. Running tens of thousands of agents in parallel requires coordinating compute, managing task dependencies, handling failures gracefully, and synthesizing outputs into coherent results. It is an operational challenge that looks less like traditional software development and more like managing a large distributed system, except the workers are AI models rather than servers. Claude Code now supports dynamic workflows that can scale coding jobs to 1,000 parallel agents, and Cherny's comments suggest real-world usage has already pushed far beyond that benchmark.
On some days I'm managing tens of thousands of agents at once.Boris Cherny, Claude Code creator, via Fortune
What This Means for Software Development
The broader implication is that the gap between how AI is discussed in product announcements and how it is actually being used in practice may be wider than most observers realize. Cherny is not describing a controlled demo. He is describing a regular working day. That raises practical questions about tooling, oversight, and how engineers stay meaningfully in the loop when the number of concurrent processes exceeds what any person can monitor directly. Cherny has spoken publicly about what this shift means for the future of software engineers as a profession, arguing that human roles will evolve rather than disappear, but that the nature of the work is changing fundamentally.
Anthropic has been building infrastructure to support this kind of usage. Features like the /goal command that lets agents work unattended for hours point to a deliberate product strategy around long-running, minimally supervised workflows. Cherny's day-to-day experience appears to be both a test case and a proof of concept for where the company believes professional software work is heading.
The conversation also touches on a broader tension in AI development: building systems capable of operating at massive scale while maintaining enough human oversight to catch errors, prevent waste, and ensure outputs are actually useful. Managing tens of thousands of agents is only valuable if the results can be reviewed, validated, and acted upon. How engineers build those review layers is becoming one of the more pressing practical problems in the field.
For anyone following the evolution of agentic AI tools, Cherny's comments are a useful data point. The technology is advancing quickly, and the people building it are already using it at a scale that would have seemed implausible just a year or two ago. Whether that pace of adoption spreads across the broader software industry, or remains concentrated among teams with direct access to the most capable models, is a question the coming months will begin to answer.
