Now that coding agents are handling nearly all coding implementation, a new challenge has emerged: keeping track of all your tasks, which agent is working on what, and maintaining overall visibility into your workflow.
Task management itself has become a significant bottleneck—ironically because we're completing so many tasks simultaneously. While it's a good problem to have (reflecting dramatically increased output), engineers must identify and resolve these bottlenecks to unlock even greater productivity.
In this article, I'll walk through my coding workflows, how I track all tasks, and how I monitor which agent is doing what work.
Why Task Organization Is Harder With Coding Agents
Understanding the "why" is essential. Task organization is more challenging today simply because coding agents enable programmers to perform far more work than before.
Previously, tasks like these could take an entire day (or several):
- Implementing a design
- Resolving a complex bug
- Implementing a new feature
Now, these are tasks you can work on in parallel and complete within a day at most. Few coding tasks today require multiple days to implement—only the most complex features.
This is a stark contrast to the pre-LLM era (before ChatGPT's release in December 2022), where you had to work on one task at a time, sequentially, and each task took significantly longer due to:
- Researching new frameworks
- Introducing and fixing unintended bugs
- Rebasing on top of rapidly changing codebases
These are essentially non-issues now because:
- Coding agents rarely introduce silly bugs anymore. Both writing and reviewing code with AI agents has virtually eliminated such errors.
- You don't need to manually learn new frameworks. Agents either already know established frameworks or can research and understand them autonomously online.
As a result, we're handling many more tasks simultaneously and need robust control over what we're working on, how to test it, and which agent is handling each task—so we can follow up effectively.
How I Approach Task Management With Coding Agents
Here's a high-level, task-agnostic overview of how I work with coding agents and keep track of everything. In the next section, I'll cover specific tools I've used and my feedback on them.
Key principles:
- Automate task tracking via API or MCP. Your task management system should connect directly to your coding agent through an API or Model Context Protocol (MCP). Do not manually write comments, update progress, or interact with the system yourself. That's a waste of time—your coding agent should handle this automatically.
- Work in separate worktrees. Always isolate work into separate worktrees. Without this, coding agents will overlap and create conflicts. Claude Code offers a built-in worktree feature, but there are many other approaches. Using separate worktrees prevents merge conflicts and keeps agent outputs cleanly separated.
This article is continued in the full version, which covers specific tools, agent assignment strategies, and testing workflows.