AI Agents vs. Human Developers: Is Delegation Really So Different?

Some engineers still reject AI agents in their daily work. Why is that?

The usual reasons I keep hearing include:

• AI agents are not good enough (yet).
• They are not predictable.
• They have no lasting memory.
• etc.

Let’s leave love of the craft and ideological reasons aside for a moment.

A small thought experiment:

Today, I distribute tasks among a large number of AI agents and then assess the results. At some point, changes are released, and I carry the responsibility. (If I had infinite time and energy, I would prefer to do everything myself.)

In another reality, these AI agents are human developers instead. I distribute tasks and assess the work. At some point, changes are released, and I carry the responsibility. (If I had infinite time and energy, I would prefer to do everything myself.)

In my view, it is at least not obvious that handing work off to a machine instead of to a human is fundamentally different. People keep saying: “Well, humans are more predictable and you can discuss things with them better.” I am not so sure that statement is broadly true. Is that really the case? Can you generalize it like that? Other people say: “The AI agent has no memory of past decisions and keeps reinventing the wheel.” Again: I am not sure human developers fundamentally do that differently. Still others are bothered by hallucinations: “The confidence with which it is wrong!” As a human, I am not always better at that.

Other people argue that software development is no longer what it used to be.

Again:

1. The work is not getting any less, and AI agents apparently are not stealing devs’ jobs after all.
2. The backlog of your software was already practically endless before; the endless work is just getting done faster now.
3. AI agents deliver better results when the software follows industry best practices. Humans do too.
4. AI code review by no means finds every bug and produces false positives. Human code review does too.

I am not saying that AI agents can or should replace human collaboration. Humans need humans, and I still firmly believe that the best results come when you bring smart people into a room to solve a problem creatively.

I am only saying that many of the points some engineers object to in Claude Code and Codex also occur with humans.

And maybe that is simply okay.