On the Human in the SDLC

The other day I watched this talk by Peter, the OpenClaw guy, and among other things he expressed a fairly interesting thought. I want to leave it here too, especially because it overlaps with my own. I had just been looking at it from a different angle before.

He argues that a human is still needed in the SDLC because product development is not a straight road uphill, but a rather chaotic movement from side to side. While we move back and forth, we learn about new system constraints or clarify the wording of the problem we are solving. Sometimes, during the process, it becomes clear that the direction we were moving in is not actually that interesting, or was simply wrong.

Agents are great at going straight uphill. A human — preferably one with taste, experience, and a sense of beauty — can redirect them toward the place we should actually be going.

On the one hand, this thought sounds a little naive in a world where we are trying to hand more and more stages over to agents. But it resonates with me. Everything still comes down to the fact that good products are created by people who care and who actually know something. How to develop those qualities is an open question. And not a question for agents.

In my own projects I can see very clearly how much has to be reworked only after I start using a tool myself as an end user. It was impossible to foresee that at the spec stage — those ideas simply did not occur to me then.