Harness 2 for Framer Agents Starts Building While It Thinks.

Harness 2 for Framer Agents lets the agent apply changes while it is still working through a prompt. Builders see progress sooner, while the agent can complete larger workflows in one run.

The rebuilt harness can read a Framer document, make decisions and apply changes while the model writes its next steps.

That change reduces the machinery behind each task.

Long-task speed

20% faster

Cost

Up to 33% cheaper

Tool steps

2x fewer

Cached input tokens

39% fewer per message

Source: Read the official announcement.

The agent no longer has to finish planning an entire response before changes appear in the document. Execution happens as the agent code streams, so completed steps can take effect while later steps are still being generated.

For a multi-part site request, this means the agent can inspect the document, decide what needs changing and begin applying those changes within the same program. The result is earlier visible progress and fewer tool steps between the prompt and the finished work.

Harness 2 also brings the same capabilities to external agents through Framer MCP. That builds on Framer’s existing support for agents that can work with sites outside the main editor, covered in Framer MCP Lets AI Agents Build and Publish Sites. The new harness changes how those agents execute the work, not just how they connect to Framer.

What is Harness 2 for Framer Agents?

Harness 2 is the rebuilt execution system behind Framer Agent. It lets the agent read a document, make decisions and apply changes while the model is still generating later steps.

How much faster is Harness 2?

Framer reports that long tasks are 20% faster. The new approach also uses 2x fewer tool steps and 39% fewer cached input tokens per message.

Which Framer plan includes Harness 2?

The announcement does not list plan availability or pricing. It presents Harness 2 as the rebuilt harness for Framer Agent and says the same capabilities extend to external agents through MCP.