Project management has been fundamentally transformed by the rise of large language models (LLMs). The way we approach product development and engineering tasks has shifted dramatically, reshaping the role of the software engineer and demanding new strategies for efficiency.
In this article, I’ll explore how to perform effective project management in the age of AI. I’ll cover both the high-level orchestration of tasks—deciding what can and should be done—and the practical execution of those tasks, ensuring you get the most out of AI tools while maintaining high-quality output.
Why Project Management Has Changed
The fundamental reason project management has evolved in the era of LLMs is the radical shift in how time is allocated across tasks in a tech project. The traditional distribution of a software engineer’s time—say, 70% writing code, 15% in meetings, and 15% testing—has been upended.
With LLMs, the time spent writing code has effectively plummeted, often to near zero for those who delegate coding to AI agents. Instead, that time is reinvested in new activities:
- 30% – Prompting agents (directing AI to write code)
- 10% – Meetings (unchanged)
- 30% – Testing (now a much larger focus, as AI-generated code requires rigorous verification)
- 30% – Additional time that can be spent on other valuable activities, such as learning, spinning up more agents, expanding project scope, or improving platform robustness
The 30% extra time is, of course, an estimate, but it represents a real opportunity. The key takeaway: project management has changed because what we spend time on has changed. This necessitates new optimizations to maximize your effectiveness—which is the core focus of this article.
When I refer to project management, I mean the time and attention you, as a software engineer, devote to working on your projects, deciding which tasks to prioritize, and how to execute them efficiently. The goal is to leverage AI to complete more work with greater speed without sacrificing quality.
Effective Project Management Strategies with AI
In this section, I’ll outline specific techniques I apply to maximize productivity in the AI era. These strategies are designed to get the most out of LLMs and their agents, allowing you to tackle more projects and deliver results faster. While these methods are based on my experience, they can be generalized and adapted to your unique workflow.
1. Map Out Work More Beforehand
The first—and perhaps most impactful—change is to invest more time in upfront planning. Tasks often arrive via Slack messages (e.g., product feedback, bug reports) or from your own project backlog. Before diving in, I make it a priority to map out the work thoroughly.
Why this matters now: a well-defined plan allows AI agents to run autonomously for extended periods. When the scope is clear, you don’t need to constantly intervene, answer clarifying questions, or course-correct mid-task. This translates into fewer interruptions and faster completion.
Consider a task with a vague or ambiguous specification—one full of unknowns. If you fire off that task to an agent without any mapping, you’ll likely hit a wall: the agent may misinterpret requirements, produce irrelevant code, or stall. In contrast, a mapped-out task breaks down the work into clear steps, defines acceptance criteria, and anticipates potential pitfalls. This enables the agent to work independently, and you can set it loose while you focus on other priorities.
Practical tips for mapping:
- Decompose tasks: Break the work into smaller, well-defined subtasks with specific objectives.
- Define deliverables: Clearly state what “done” looks like for each subtask.
- Anticipate challenges: Identify potential technical or design obstacles in advance, and provide guidance on how to handle them.
- Set boundaries: Specify the scope and any constraints (e.g., APIs to use, files to modify, performance targets).
By spending 80% of your time on planning and 20% on execution, you’ll find that agents can handle the heavy lifting, leaving you with more time to scale your efforts across multiple projects.