Approval moves to the beginning
Manus introduced Plan Mode on July 22 as an optional step before execution. When activated, the agent checks feasibility, asks for missing context and produces a structured Markdown plan. The user can edit that document, request revisions and decide when the agent is allowed to proceed.
This is a small interface change with a large governance effect. Review after execution can catch an error, but it cannot recover time already spent on the wrong architecture or audience. Reviewing the approach first makes assumptions visible while they are still cheap to change.
Useful beyond software development
The company emphasizes development examples, where a plan can expose a missing integration, flawed schema or misunderstood requirement before code is written. The same pattern applies to research, presentations and video: agree on sources, audience, structure and output constraints before generation begins.
Plan Mode can also be enabled during an active task. That creates a natural stage gate between exploration and execution, or between a first deliverable and a riskier second phase. It is especially useful when the requested change touches a live system.
What a good agent plan should contain
A plan is only valuable if it is specific enough to review. It should identify the goal, inputs, output, affected systems, assumptions, validation method and any action that requires approval. A generic list of steps can create confidence without exposing the decisions that matter.
Teams should save the approved plan with the resulting artifact and compare what happened with what was promised. That record supports debugging and helps determine whether the agent is becoming more dependable or merely more persuasive.
Sources & further reading
Social-media activity is treated as a signal of attention, not proof. Product claims are attributed to the linked publisher or announcement.