Rebuilding Manus with Manus: what changed, what broke, and how we work today
Manus's co-founder describes how engineering, product and marketing teams work with agents, including what failed. People decide and own the outcome while agents do the work.
What the article says
Tao Zhang, co-founder of Manus, documents an internal experiment from the final stretch of Manus 2.0: people and agents working side by side by default. An engineer now commonly juggles 10 to 20 requests at once. The agent reads a request, finds the relevant code and proposes a change, while the human steers, checks and owns the result.
The core tool is Mono, a single entry point with a handbook written for the agent, covering every repository, a library of architecture decisions and end-to-end tests. Agents triage bugs, fix and verify them, but nothing is merged or released without a person's approval. A fully autonomous loop left more than 200 items awaiting a decision after one weekend, showing that autonomy needs a verifiable goal. The bottleneck moved to deciding what to build, and about half of the PMs now ship features themselves. The post also admits fragmented attention, scattered information and permissions that need rethinking.
Read the official article AnnuIA profile
Summary written by AnnuIA in its own words from the official page of Manus, filiale de Meta. The content and rights belong to the publisher: please read the original article.