How Boris Cherny Runs 2,000 Overnight AI Agents From His Smartphone

Boris Cherny, the creator of Claude Code, has revealed details of how he manages AI agents. Every night, he says, he launches about two thousand autonomous programs that write code, test it and improve the product on their own.

The scale of automation

Cherny built a system that lets him manage a huge number of AI agents with minimal attention. All of these agents work at night while developers sleep, handling routine tasks of writing and refactoring code.

By his account, he manages this army mostly from his smartphone: he sends commands, tracks progress and adjusts the direction of the work for the next day. This is not just a demonstration of Claude's technical capabilities but a practical workflow that works in the real world.

Cherny uses Claude Code as a platform for building his own workflow. The agents run in parallel, independently of each other, and complete their assigned tasks without human intervention in the intermediate steps.

What distinguishes his approach is that quality control stays in his hands. In the morning, Cherny reviews the agents' output, accepts or rejects changes, and gives instructions for the next cycle. It is like managing a remote team, but at the scale of thousands of simultaneous "employees."

How it works in practice

This approach shows what software development may look like in the near future. Instead of hiring more programmers for routine work, a company can scale through automation. AI agents take over repetitive, predictable work:

• Refactoring existing code and improving readability
• Writing boilerplate code and standard components
• Finding and fixing potential bugs
• Optimizing performance and memory use
• Generating unit tests and technical documentation

Humans still keep control: Cherny monitors quality, checks the logic of changes and steers the agents in the right direction. This is critically important because AI can make mistakes in architectural decisions or miss nuances of business requirements. The developer acts as an architect and quality integrator, not as an executor.

Scaling without hiring new developers

For startups and companies, this makes obvious economic sense. Instead of hiring junior developers and spending months training them, you can give a tool to experienced specialists and let them focus on complex problems. AI will handle the simple, repetitive work.

This is not about layoffs. History shows the opposite: tools expand a team's capabilities, letting fewer people do more. A developer equipped with 2,000 automated agents can maintain a project that used to require a team of 20-30 people.

But there is an important caveat: this works only because an experienced developer stands behind the system, one who knows which tasks to automate, how to phrase them properly for the AI and what counts as an acceptable result. Without that experience, the system will simply start generating low-quality code.

What this means

Cherny's story is not about AI replacing developers. It is about a quality tool in a professional's hands turning them into a developer with superpowers. Two thousand overnight agents are an economy of scale, not a replacement for labor.

The previous generation of tools (IDEs, frameworks, the cloud) already did the same thing. One specialist can now do work that 20 years ago required a small team. AI is the next step in the evolution of development tools.

For companies, the takeaway is simple: invest in tools for your best developers, not in the number of developers.