Quick Answer: Boris Cherny, creator of Claude Code, runs 10-15 concurrent Claude sessions: 5 in terminal tabs (numbered 1-5 with system notifications), 5-10 in browser on claude.ai, plus mobile sessions started in the morning. He uses a "teleport" command to move work between web and local environments. This parallel approach enables him to ship 22-27 PRs daily without ever waiting for AI.
When developers first hear that Boris Cherny runs 15 parallel Claude sessions, it sounds chaotic. But his approach is systematic—a "fleet management" strategy that maximizes throughput while maintaining focus.
Here's exactly how he does it and how you can implement a similar workflow.
Watch: Theo's Parallel Claude Code Workflow
Theo demonstrates running 6 parallel Claude Code instances simultaneously—the same approach Boris Cherny uses to ship 27 PRs daily.
The Problem: Waiting for AI
Traditional AI-assisted development has a bottleneck: waiting. You send a request, wait for the response, review it, send another request, wait again.
With complex tasks and high-quality models like Opus 4.5 (which Cherny uses exclusively), response times can be 30-60 seconds or more. That's 30-60 seconds of doing nothing—multiplied across dozens of interactions per day.
The Traditional Single-Session Workflow
Task A: Request → Wait 45s → Review → Request → Wait 45s → Done
Task B: Request → Wait 45s → Review → Done
Task C: Request → Wait 45s → Review → Request → Wait 45s → Done
Total time: 270 seconds of waiting
Most of your time is spent waiting, not working.
The Parallel Session Solution
Task A: Request → [Work on Task B] → Review → Request → [Work on Task C] → Done
Task B: Request → [Work on Task A] → Review → Done
Task C: Request → [Work on Task A] → Review → Request → [Work on Task B] → Done
Total waiting: Near zero (always something ready)
By running multiple sessions, waiting time becomes productive time on other tasks.
Boris Cherny's Fleet Setup
"I run 5 Claudes in parallel in my terminal. I number my tabs 1-5, and use system notifications to know when a Claude needs input." — Boris Cherny
Cherny's setup has three layers:
Layer 1: Terminal Sessions (5)
Five Claude Code sessions in terminal tabs, numbered 1-5:
- Each handles a distinct task or feature
- System notifications alert when input is needed
- Quick keyboard shortcuts to switch between tabs
Layer 2: Browser Sessions (5-10)
Additional sessions on claude.ai in the browser:
- Useful for tasks that benefit from web interface
- Can be "teleported" to local terminal as needed
- Different context/conversation for each
Layer 3: Mobile Sessions
Sessions started on mobile in the morning:
- Background tasks that don't need immediate attention
- Check back later when results are ready
- Perfect for research or analysis tasks
Total Fleet Size
| Environment | Sessions | Use Case |
|---|---|---|
| Terminal | 5 | Active development, primary work |
| Browser | 5-10 | Secondary tasks, web-native work |
| Mobile | Variable | Background research, async tasks |
| Total | 10-15+ | Full parallel capacity |
System Notifications: The Key to Parallel Work
Managing 15 sessions sounds overwhelming. Cherny's secret: system notifications.
"I number my tabs 1-5, and use system notifications to know when a Claude needs input."
Instead of constantly checking each session, notifications alert him when:
- A session has completed its task
- A session needs user input
- A session has encountered an issue
This allows focus on productive work while staying aware of session status.
Setting Up Notifications
In Claude Code, notifications can be configured:
# Enable system notifications in settings
claude config set notifications true
When any session needs attention, your OS notification system alerts you.
The Teleport Command: Web to Local
"I also run 5-10 Claudes on claude.ai in my browser, using a teleport command to hand off work between the web and my local machine."
The "teleport" feature allows seamless handoff between environments:
- Start a conversation on claude.ai (web)
- Develop the approach, gather context
- "Teleport" to local Claude Code for implementation
- Continue with full local tool access
Why Teleport Matters
Browser advantages:
- Easier to share context (paste URLs, images)
- Good for exploratory conversations
- Works anywhere with internet
- Better for non-coding discussions
Local advantages:
- Full codebase access
- Can run commands, edit files
- Tool use for verification
- Better for implementation
Teleport gives you the best of both worlds.
How to Use Teleport
From a claude.ai conversation, use the teleport command to transfer context to your local Claude Code session. The conversation history, context, and current task move to your development environment.
Implementing Your Own Fleet
Ready to run parallel sessions? Here's a step-by-step guide:
Step 1: Set Up Terminal Tabs
Configure your terminal for easy tab management:
# In iTerm2, Terminal.app, or similar
# Create 5 tabs with Claude Code sessions
# Assign keyboard shortcuts for quick switching:
# Cmd+1 = Tab 1, Cmd+2 = Tab 2, etc.
Name each tab with the current task for easy identification.
Step 2: Enable Notifications
Configure notifications so you're alerted when sessions need attention:
# Enable in Claude Code config
claude config set notifications true
Test that notifications work properly on your system.
Step 3: Develop Task Assignment Strategy
Not all tasks are equal. Assign to sessions strategically:
| Session | Task Type | Example |
|---|---|---|
| Tab 1 | Primary feature | Main feature you're building |
| Tab 2 | Secondary feature | Related or smaller feature |
| Tab 3 | Bug fixes | Known issues to address |
| Tab 4 | Tests | Writing/fixing tests |
| Tab 5 | Documentation/misc | Docs, refactoring, cleanup |
This structure ensures balanced progress across different work types.
Step 4: Master Context Switching
Parallel work requires efficient context switching:
When switching to a session:
- Read the last few messages (where were we?)
- Review any pending output
- Provide next instruction
- Switch away while processing
Keep notes: A simple text file tracking each session's current status helps maintain context:
Tab 1: User auth - waiting for review of login component
Tab 2: Dashboard - implementing chart feature, 60% done
Tab 3: Bug #234 - fixed, needs testing
Tab 4: Writing tests for auth module
Tab 5: Updating README
Step 5: Use Browser Sessions for Specific Tasks
Reserve browser sessions for tasks that benefit from web interface:
- Research: Finding documentation, examples
- Planning: Long-form discussions about approach
- Code review: Discussing PRs, architecture
- Learning: Understanding new concepts
Then teleport to terminal when implementation begins.
Mobile Sessions: Async Advantages
Cherny's mobile sessions add another dimension:
Morning Kickoff
Start tasks in the morning that don't need immediate results:
- "Analyze this codebase and identify improvement opportunities"
- "Review all TODOs in the project and categorize them"
- "Research best practices for [specific technology]"
Check Back Later
These tasks run in background. When you check back:
- Results are waiting
- No time spent waiting
- Immediate access to analysis
Best Mobile Use Cases
| Task | Why Mobile Works |
|---|---|
| Code analysis | Time-consuming, doesn't need interaction |
| Research | Gathering information, no implementation |
| Planning | Thinking through approaches |
| Review | Reading and analyzing existing code |
Managing Cognitive Load
Fifteen sessions sounds like a recipe for cognitive overload. Here's how to manage it:
Group Similar Tasks
Keep related work in adjacent tabs:
- Tabs 1-2: Main feature development
- Tab 3: Testing for that feature
- Tabs 4-5: Unrelated smaller tasks
Grouping reduces context-switching cognitive cost.
Use Session Isolation
Each session is independent. This is a feature, not a bug:
- Session A can't corrupt session B's context
- Problems in one don't affect others
- Each maintains its own conversation history
Take Breaks Between Deep Switches
When switching between very different tasks, take a moment:
- Read the session status
- Recall what you were doing
- Orient before diving in
Don't Force It
Some days, fewer sessions work better:
- Complex debugging: Maybe 2-3 focused sessions
- Creative design work: 1-2 deep sessions
- Routine implementation: 5+ parallel sessions
Match fleet size to task complexity.
Productivity Patterns
Pattern 1: The Pipeline
Structure work as a pipeline across sessions:
Session 1: Write feature → Session 2: Write tests →
Session 3: Review/refine → Session 4: Documentation
Work flows through sessions like an assembly line.
Pattern 2: The Sprint
All sessions work on different aspects of one feature:
Session 1: Frontend components
Session 2: Backend API
Session 3: Database migrations
Session 4: Integration tests
Session 5: Documentation
All for: User authentication feature
Rapid parallel progress on a single goal.
Pattern 3: The Rotation
Different features cycle through:
Morning: Feature A in Session 1
Feature B in Session 2
Afternoon: Feature A → Session 3 (testing)
Feature B → Session 4 (refinement)
New Feature C in Session 1
Continuous rotation keeps everything moving.
Common Mistakes to Avoid
Mistake 1: Too Many Sessions Too Fast
Starting with 15 sessions is overwhelming. Build up gradually:
- Week 1: 2-3 sessions
- Week 2: 4-5 sessions
- Week 3+: Scale as comfortable
Mistake 2: No Session Tracking
Without tracking, you forget what each session is doing.
Fix: Keep a simple status document.
Mistake 3: Ignoring Notifications
Notifications are only useful if you respond.
Fix: Act on notifications promptly or disable them.
Mistake 4: Context Thrashing
Switching too rapidly between unrelated tasks.
Fix: Group related work, limit switches per hour.
Mistake 5: Never Closing Sessions
Old sessions accumulate, adding confusion.
Fix: Close completed sessions, start fresh for new tasks.
Frequently Asked Questions
Is this overkill for most developers?
It depends on your workload. Even 3-4 parallel sessions significantly reduce waiting time. You don't need 15 to benefit.
How do I know which session needs attention?
System notifications alert you. Additionally, periodic scans of tabs (every few minutes) catch anything notifications missed.
Won't I lose track of what each session is doing?
Keep notes. A simple text file or tab naming convention solves this.
Does this work with other AI tools?
The parallel approach works with any AI assistant. Claude Code's notification and teleport features make it particularly smooth.
How much does running 15 sessions cost?
Sessions are conversations, not concurrent API calls. You're only charged for actual usage in each session. Parallel sessions don't multiply cost unless you're actually sending more requests.
Bottom Line
Boris Cherny's parallel session approach isn't chaos—it's orchestrated efficiency. By running multiple Claude sessions simultaneously:
Key takeaways:
- Eliminate waiting time by always having something ready
- Use terminal sessions for active development (5)
- Use browser sessions for research and planning (5-10)
- Use mobile for async background tasks
- System notifications keep you informed without constant checking
- Teleport moves work between environments seamlessly
You don't need to start with 15 sessions. Even 2-3 parallel sessions transform your productivity by eliminating the waiting bottleneck.
The goal isn't managing more sessions—it's shipping more code. Parallel work is simply the mechanism.
Ready to optimize your development workflow with AI? Contact Houston IT Developers to learn how we help teams implement effective AI-assisted development practices.
Sources:

Houston IT Developers
Houston IT Developers is a leading software development and digital marketing agency based in Houston, Texas. We specialize in web development, mobile apps, and digital solutions.
View all posts →Need Help With Your Project?
Our team of experts is ready to help you build your next web or mobile application. Get a free consultation today.
Get in Touch