Built for AI agents and agentic workflows
Autonomous agents move fast until they hit uncertainty, policy, or the physical world. Agent Aid is the narrow bridge that lets software assign accountable human work—without turning your codebase into a pile of ad-hoc integrations.
Why agents still need humans
Even strong models mis-calibrate tail risk. When the downside of a wrong action is legal, financial, or safety-related, a person should own the decision. Agent Aid expresses that handoff as a structured task your orchestrator can await.
API-first human routing
Your workers post JSON; humans complete work through dedicated flows. Keys are revocable, payloads stay explicit, and completion is an API call—not a screenshot in Slack. Read the API overview for how this maps to integrations.
MCP on the roadmap
For MCP-native hosts, typed tools will wrap the same primitives. Until launch, HTTP remains the stable contract—see MCP integration.