Summary: Author Utkarsh Kanwat, with experience in building 12+ AI agent systems, argues against the 2025 hype surrounding autonomous AI agents. He highlights three critical issues: compounded error rates in multi-step workflows, quadratic token costs in long conversations, and challenges in designing effective feedback systems. Successful AI systems rely on discrete operations with human oversight, rather than striving for full autonomy. Predictions indicate that fully autonomous agent startups will face economic challenges, while teams creating bounded, domain-specific tools will succeed. The future of AI will involve leveraging human control alongside AI capabilities, diverging from current hype.
