# Running agents in parallel can be a total chaos. Learn how we avoid it, at Latent AI.
One agent is a worker. Many is a team.
Boris Cherny, the creator of Claude Code, ships 20-30 PRs a day by running 5 Claude instances in parallel - roughly 3x the throughput of a single session. If you're running Claude Code one task at a time, you're not maximizing the potential of working with agents.
In this session you will see how we run agents in parallel in our day-to-day coding workflows at Latent AI \- and how we delegate the noisy legwork to subagents \- with short live demos\.
This session is a taste of our full AI Coding Agents training we've built for R&D teams. For more info, reach out to Danel at danel@reachlatent.com.
📢 This session will be conducted in Hebrew.
🚀 What we'll cover
Parallel agents in Claude Code \- what can actually run in parallel\, one worktree per task\, and the merge flow that keeps you out of conflicts
Subagents \- what they are\, why they beat one bloated context \("context rot"\)\, built\-ins vs custom agents\, and when to use a subagent vs a Skill
Patterns that work, and best practices.
✨ You should attend if
You use Claude Code and feel like you spend more time waiting than working
You've heard about subagents or worktrees but haven't made them part of your flow
You want to see how a real team runs multiple agents without chaos
📌 Details
Language: Hebrew
Host: Jonathan Yarkoni \- Co\-founder & CEO of Latent AI
🙋 We'll finish with an open Q&A - bring your questions.
About us: Latent AI is an AI engineering firm that has built hundreds of production AI agents and machine learning solutions for enterprises. Our Academia was built on that hands-on expertise, with one mission: making sure no one is left behind in AI. We've trained dozens of R&D and business teams around the world on how to utilize the potential of Cursor, Claude Code, Cowork and many more tools.
See you there!