I am really fascinated and intrigued about what agentic systems could offer —AI agents that can independently tackle complex tasks. I’m skeptical about how practical these systems are right now. My own experience with them lacks the consistency needed for reliable use. That said, I would still love for AI to reach a point where it can start finding its own solutions to the problems I throw at it, like summarizing a webpage and forwarding it to a project team without me having to micromanage the steps. An reading the news, this is were a lot of research and betas are heading right now.
But now, I came across an interesting quote: “The point is building these complex agents has been proven a waste of time over and over again... It's much easier to swap in a single API call and modify one or two prompts than to rework a convoluted agentic approach... the same prompts can't be reused reliably between different models.”
It made me think that, for now, it might be more practical to focus on simpler solutions. Instead of trying to build fully autonomous agents, we might be better off creating targeted workflows that string together specific instructions. In reality, I can imagine that 20 workflows would cover 80% of my routine tasks.
This approach seems more realistic in the short term—structured and manageable, without the headache of inconsistent agent behavior. It’s a reminder that progress doesn’t always come from chasing the most complex solution, but from refining what works.
