TanStack AI Brings Typed Outputs To Harnesses

This is a meaningful infrastructure upgrade: structured output becomes useful at the end of real agent runs, not just simple one-shot prompts.

• –Agent loops can execute tools first, then produce a schema-validated final object
• –Harness adapters support agents including Claude Code, Codex, OpenCode, Grok Build, and ACP-compatible agents
• –TypeScript inference spans schemas, tool calls, streaming events, and final results
• –The feature makes agent outputs easier to feed into UIs, workflows, persistence layers, and automated tests
• –Provider portability remains a core advantage, though validation and streaming behavior vary by execution path