The harness war is already over

LangChain launched Deep Agents Deploy in beta.

Claude launched Managed Agents.

Agno has AgentOS.

All of these frameworks converge on the same agent architecture, which is becoming the standard, with the ability to create agents (harnesses) simply.

Building autonomous agents has never been easier. That is only the start of the adventure.

The shared skeleton

All of these frameworks converge on this architecture:

• plan
• tools
• memory
• reflection loop

All of that can now be written in a few lines with the frameworks above.

I had started working on a no-code agent builder myself, because their architecture is more or less always the same.

Only the task and the tools change.

For the curious: the experiment is linked from the original LinkedIn post. There is a strong chance it stays an experiment — everything moves too fast.

What actually matters now

What matters now is the architecture behind it and observability.

Making an agent became easy. Monitoring it and making sure it still works over time is much less simple.

The harness became a commodity. So you will be sold the infrastructure around it: servers, databases, observability, and so on.

Or you can choose, as I tried with my personal project, to build it in-house. That is not as simple as it looks.

In reality, the harness war is already over.

Now everything is about infrastructure and the privacy of your data.