I Ran 10+ AI Coding Agents in Parallel. The Bottleneck Wasn't the AI.

<p>AI coding agents changed the way I work.</p> <p>When I started using them, the workflow was simple:</p> <blockquote> <p>Give an agent a task → wait → review the result → continue.</p> </blockquote> <p>Running one agent was easy enough.</p> <p>Then I started running multiple agents in parallel.</p> <p>At first, it felt like a huge productivity boost. Different agents could handle different tasks at the same time:</p> <ul> <li>1. one agent implementing a feature</li> <li>2. another fixing bugs</li> <li>3. another exploring a technical approach</li> <li>4. another reviewing code</li> <li>5. another handling repetitive tasks</li> </ul> <p>The limiting factor no longer seemed to be how fast I could write code.</p> <p>Then something unexpected happened.</p> <p>The bottleneck became me.</p> <p><strong>My current setup</strong></p> <p>My workflow usually involves more than 10 agents running at the same time.</p> <p>They are not all the same type of agent:</p> <ul> <li>1. Claude Code</li> <li>2. Codex</li> <li>3. Kimi</li> <li>4. other task-specific agents</li> </ul> <p>More than half of them are coding assistants, which means they consume a significant amount of CPU and memory.</p> <p>Running everything on one machine is not practical, so I spread them across multiple machines and manage them remotely through terminal sessions.</p> <p>On paper, this sounds like a good setup.</p> <p>In reality, it creates a different problem.</p> <p><strong>The problem is not running agents. It is knowing what is happening.</strong></p> <p>When you have many agents running, the first question is no longer:</p> <blockquote> <p>"Can this agent finish the task?"</p> </blockquote> <p>The question becomes:</p> <blockquote> <p>"What is every agent doing right now?"</p> </blockquote> <p>I started noticing that I was spending more and more time checking sessions instead of making decisions.</p> <p>Questions like:</p> <ul> <li>1. Is this agent still working?</li> <li>2. Did it finish already?</li> <li>3. Is it waiting for my input?</li> <li>4. Did it get stuck?</li> <li>5. Is this session still relevant?</li> </ul> <p>The agents were running, but my visibility was getting worse.</p> <p><strong>Context switching becomes the real cost</strong></p> <p>The biggest mental overhead comes from switching between sessions.</p> <p>Every time I open another terminal tab, I need to reconstruct the context:</p> <ul> <li>1. What was this agent working on?</li> <li>2. Why did I start this task?</li> <li>3. What decisions were already made?</li> <li>4. What should happen next?</li> </ul> <p>This is manageable with one or two agents.</p> <p>With ten or more, it becomes exhausting.</p> <p>The problem is not that the agents are not capable.</p> <p>The problem is that humans are still limited by attention.</p> <p><strong>Multiple machines make it harder</strong></p> <p>Because coding agents are resource-heavy, many people cannot simply open 10 agents on one laptop.</p> <p>My workflow requires multiple machines.</p> <p>Remote terminal tools work well for connecting to those machines, but they mostly show the terminal itself.</p> <p>They don't tell me:</p> <ul> <li>1. which sessions are active</li> <li>2. which ones are waiting</li> <li>3. which ones finished</li> <li>4. which ones need attention</li> </ul> <p>So the only way to know is manually checking each tab.</p> <p>Ten agents means ten context switches.</p> <p><strong>tmux solves a different problem</strong></p> <p>I have tried tmux and other terminal multiplexers.</p> <p>They are excellent tools.</p> <p>They solve an important problem:</p> <blockquote> <p>How do I keep my sessions alive and organized?</p> </blockquote> <p>But managing multiple AI agents introduces another problem:</p> <blockquote> <p>How do I manage my attention across many autonomous processes?</p> </blockquote> <p>Keeping everything visible is not always better.</p> <p>A screen full of active sessions ca