At Waves '26, Make unveils MakeOS and a set of controls for running AI agents in plain sight
At its Waves '26 conference, Make CEO Fabian Veit laid out a "glass box" idea for the agent era: every action, by a person or an agent, should be visible, permissioned and reversible. The announcements fall into four areas. In building, agent orchestration, where an agent hands jobs to subagents, is available now, while pausing a scenario for human approval from Slack, Teams or the Make mobile app, and pinning test data on a module, are coming soon. Agents also get an execution layer: Functions for code the agent runs, Endpoints that grant access to a single API action instead of a whole app, Knowledge as a shared source of context (available), and MCP toolboxes that define exactly what each agent can use and work with Claude, ChatGPT and any agent that speaks MCP. In governance, Custom Roles and Private Spaces are available, and an organization policy enforced at runtime is on the way. In observability, the Make Grid map, event subscriptions and an audit log covering agent and MCP actions are available, with agent evals and AI spend tracking coming soon. On top of this, CTO and co-founder Patrik Simek introduced MakeOS, a new product that shows every agent, what it is working on, what it costs and who owns it, and that blocks an agent going past its limits, asks a person and logs the attempt. The post gives no availability date for MakeOS.
Otomatzia editorial · Published
Source published: · Make
