This is not one of those posts that tries to convince you how bad coding agents are. I am very much pro-AI and specifically for coding agents - they are here to stay. I’ve been maxing my tokens with Vibe Coding every day. But, as I have been doing this, there have been some patterns I’ve experienced, that I wanted to share here.

As always, let’s first reminisce with some melancholy music to the good old days :) I remember staying glued to my computer screen all day to implement a simple streaming JSON parser because my file was too big. At the end of the day, I used to feel I’ve maxed out tokens from my personal organic neutral network. Even if in the schema of the whole project this would be a small factor, I would feel productive that day.

These days with agents, I am feeling that I am serving a role of ISR (Interrupt Service Routine) to these agents. I think hard and come up with a prompt, then I wait, and wait and wait for the agent to come back to me with a plan, then I review the plan and give it a go. Then, I wait, and wait, and wait for it to finish the task. Then, I decide to try it out and iterate over it while, yes, waiting.

I am now able to finish the whole project within 1-2 hours. The agent is busy 90% and me 10%. First, I thought, wow, this is going to be so productive, I can now multitask easily. But, again, the same systems principles apply here - cost of context switching. Somehow, I felt context switching with multiple agents is even harder. I read people argue - wait till we get fully autonomous agents like OpenClaw working reliably, so you can cut your 10% to only the initial prompt (fire and forget). Maybe that would help or maybe that would make the cost of context switching worse (cycle of want).

See, I am NOT saying coding agents are bad or stop using them. On the contrary, I believe they are an integral part of developers’ lives, but we need to build and design right workflows to make us (humans) more engaged and efficient with less toil. The tools and practices we are employing with these agents today might not be the best way forward.