Why Does Coding With Agents Need Different Tools?

David Heinemeier Hansson says programming with agents requires a different tool set because the work shifts from one problem at a time in your head to parallel processing. Waiting on a single agent does not feel productive, even when that one agent is highly productive. It feels a little useless.

He used TextMate for almost 20 years and thinks he started in 2005. He helped get the first version out, then was not interested in an alternative until the switch to Linux forced him out of his habitat. Writing code by hand in TextMate or even NeoVim meant focusing on one problem and working through it methodically to the end. Deep immersion in a single problem was the portal to flow. Agents are at once too fast and too slow. They do not give an immediate reply the way typing on a keyboard does, so you let the agent cook. Sitting around waiting for them does not feel productive and does not feel good. During the first agentic moment he ran mostly one agent at a time and felt unsure about it. A lot of hard problems can be solved by throwing more resources at them.

He hates the term agentic engineering. The plain name for the work is still programming, because programming with agents is still programming.