Anthropic launched a new product today.
The positioning is: You don't need to do Harness anymore, I'll do it for you. I understand the limitations of the model best, I'll design Harness according to the model's characteristics, and package it up and sell it to you. And this package is Managed Agents.

In conclusion
Previously, everyone was saying: Harness means that the truly scarce capabilities are not inside the model, but outside the model.
But today, Anthropic is trying to put this "outside the model" layer back "inside the model manufacturer".
Harness Engineering won't die. But the window of opportunity for "human-made harnesses" may be shorter than anyone anticipates. Your Harness will either be rendered obsolete by model advancements, replaced by platform services, or you'll have to outpace both. However, realistically, general-purpose Harnesses can cover most scenarios, but not all. Legal, financial, and medical agents each have their own evaluation standards and security boundaries. Achieving 80% on a general-purpose platform isn't difficult; the final 20% is where competitiveness lies. Moreover, the open-source community will explore many more interesting avenues, such as Hermes, which I shared yesterday. Let the Agent write its own Skill, use the Skill, and modify the Skill. Don't hand over Harness to the platform; let the Agent write its own Harness.