1300 166 138 Tell us what you need
News / Anthropic

Claude Code’s creator runs tens of thousands of AI agents

By Monika Tantau • Published 5 months ago
Claude Code's creator runs tens of thousands of AI agents

Key takeaways

• Boris Cherny, who built Claude Code, says some days he runs tens of thousands of AI agents at once
• Anthropic's Claude Opus 4.8 can plan and run hundreds of sub-agents in a single session
• The shift shows business leverage now comes from orchestration, not headcount
• For Australian SMBs, a small team can take on work that once needed a department

Most Read

Boris Cherny, the engineer behind Anthropic’s Claude Code, told Fortune this month that on a busy day he now oversees tens of thousands of AI agents at once. A year ago that would have read as science fiction. Today it describes a working pattern spreading from frontier labs to ordinary businesses. The number matters less than what it signals: the unit of work in software and operations is no longer a person at a keyboard. It is a person directing a fleet of agents that run on their own.

How one person runs thousands

Cherny was not describing a demo. Anthropic’s latest release, Claude Opus 4.8, ships with what the company calls Dynamic Workflows, where a single session can plan and run hundreds of sub-agents in parallel. Each agent takes a slice of a larger task, works without step-by-step supervision, and reports back. Stacked across a day, the count climbs into the thousands. The model has moved from answering one prompt at a time to managing a queue of work the way a team lead manages people.

From one chatbot to a fleet

Most businesses still picture AI as a chat window. Someone types a question, the model types back. That framing is two product generations behind. The practical shift is delegation. Instead of asking an agent for a draft, an operator hands it a job, such as reconciling a week of payments or clearing a backlog of documents, and checks the result later. The same pattern underpins back-office tools like finance and accounting automation, where the work runs while the office sleeps.

Leverage stops matching headcount

This is where the Cherny figure becomes useful rather than just impressive. For two decades, output scaled roughly with the size of a team. More work meant more hires. Agent orchestration breaks that link. A two-person business can take on work that once needed a department, not by hiring but by directing. The constraint moves from how many people you employ to how clearly you can define a task and verify what comes back.

Why Australian businesses should pay attention

For Australian operators the timing is pointed. Labour is expensive, skilled staff are hard to find, and the National AI Centre reports that the biggest barrier to adoption is not cost but trust, with around two in three non-adopters preferring to keep humans in control. Agent orchestration does not remove that control.

It changes the human role from doing the task to briefing and checking it. Melbourne and Sydney firms that learn to supervise agents well will widen the gap on those still treating AI as a search box. The sensible first step is a narrow, well-defined process and a clear AI strategy, not a wholesale handover.

Why Australian businesses should pay attention

What this means next

The headline figure will keep climbing, and the raw count is the least interesting part. The skill that now separates businesses is knowing which work to delegate, how to define it, and how to check it safely. Anthropic has already started metering agent usage across its subscriptions, a sign that running agents at scale is becoming a normal line item rather than a research novelty. For Australian businesses, the question is no longer whether one person can run a thousand agents. It is which thousand, and to what end.

Scroll to Top