AI promised us free time. Instead, agents gave us infinite context switching.

Coding is cheap now, but managing parallel AI agents is burning out engineering leaders with relentless context switching.

Tools · Source: Hacker News

What happened

Orestis Ioannou leads engineering at Bold.org. He started using AI agents for all his coding work this year. The pitch from the AI industry was simple. Write less boilerplate, delegate the boring parts, and go home early. The reality is the exact opposite. He is working harder than ever. The workload feels heavier. It is not because any single task is harder to execute. It is because there are so many more tasks happening all at once.

He pulled his GitHub contribution data from 2019 to 2026 to prove it. As he moved from Full Stack Engineer to Principal Engineer and then VP of Technology, his code contributions naturally dropped. He averaged 114 contributions a month for years as he focused on architecture and planning. Then 2026 happened. AI pushed his output to 2,102 contributions by October. August 2026 was his busiest month in his entire timeline with 427 contributions. His heavy days became routine again.

The root problem is how agents operate. Before AI, a normal week meant two or three serious tasks. He could hold one problem in his head, finish it, and move on. Now, he runs several agents in parallel. One implements a change. Another investigates a production issue. A third does research. None of them need him all the time, but every single one needs him eventually. They come back with questions, plans to approve, or code diffs to read.

Key facts

Why it matters

The fundamental bottleneck in software engineering just shifted. Writing code is no longer the hard part. Managing the output of autonomous agents is the new constraint. Founders and engineers can now spin up dozens of parallel tasks because implementation is incredibly cheap. But human attention does not scale like compute. Every time an agent pauses for human feedback, it forces a massive context switch. You have to reload a different problem, a different repository, and a different set of decisions into your brain. This drains mental energy faster than writing the code yourself.

We are entering an era of AI-induced burnout driven by endless cycles of bursts and fatigue. When implementation costs drop to zero, there is no natural friction to stop us from pursuing every single idea. Ioannou noted that his capacity to implement ideas grew much faster than his capacity to think about them. Engineering teams will need entirely new frameworks to limit work in progress. Otherwise, senior engineers and managers will simply drown in pull requests generated by their own bots. The low season for engineering work is disappearing.

For builders

Enforce strict agent WIP limits

Do not let your team run unlimited autonomous agents simultaneously. Set hard limits on work in progress. Companies that fail to do this will pay the price with burned-out senior engineers who spend all day context switching.

Build context-aware review tools

The current interface for reviewing agent code is completely broken. Founders who build tools to reduce cognitive load during agent handoffs will win big. Large engineering teams will eagerly pay a premium for this solution.

Redefine engineering productivity metrics

GitHub contributions are no longer a valid metric for human productivity. AI inflates these numbers while hiding the mental toll. Engineering leaders must measure cognitive load and context switches instead, or risk losing their best talent.

My take

We built AI to replace our keyboards, but we forgot to upgrade our brains. Infinite leverage is completely useless if it just turns you into a tired middle manager for a fleet of relentless bots. Stop starting new tasks just because you can. Learn to say no to your own agents before they burn you out.

Original reporting: Hacker News. This is my rewrite and opinion.

More AI news for builders