Manus Details AI-Native Reorganization: Agents Exposed to Full Engineering Stack

Manus described an AI-native organizational overhaul in which engineering infrastructure is rebuilt to expose repositories, end-to-end tests, production logs, alerts, email and feature flags directly to agents. Engineers run 10-20 concurrent agent sessions, leadership delegates 20-plus tasks to a master session, and seven people maintain four apps, with token spending of about $10,000 per person per month. Agent conversation sessions replace documents and dashboards as the organization's core work objects and memory.
Event version 1.
Key insights
Infrastructure exposed to agents
Manus connected scattered repositories into a unified entry point, enabled agents to run end-to-end tests in seconds, and gave agents permission to read production logs, view alerts, send email and control feature flags. The goal is for agents to see and operate everything humans can, closing a modify-verify-troubleshoot loop.
Goals and evidence replace process control
Engineers pass Slack requirement text verbatim to agents without explaining how to execute. Manus uses top models as code review bots and has agents walk through registration and payment with test accounts while recording screens, so product managers can review the recording.
Asynchronous, massively concurrent workflows
Engineers run 10-20 sessions simultaneously; leadership assigns 20-plus to-dos to a master session that pulls context from Slack, Gmail and Notion and distributes tasks. A marketing agent receives tasks at one workbench, splits them across parallel sub-sessions, wakes on scheduled signals and files requests when tools fall short.
Flattened roles and high token cost
One person writes frontend, backend and mobile code; seven people maintain four apps; all product managers have committed code and half can ship features independently. Manus spends about $10,000 per person per month on tokens, described as buying time, attention and reduced communication overhead.
Sessions as organizational assets
Conversation sessions and threads are replacing documents and dashboards as core work objects and memory. Operations staff wrote the release process into Agents.md so anyone can have AI release, and project members can view each other's sessions instead of asking for progress.