Everything today decides what an agent can do.
This work demonstrates how we decide whether it should.
IntentGate is a working demonstration of a missing control point in AI agent authorization: the moment between delegation and execution. Existing systems determine what an agent can do. This work evaluates whether it should still do it at the moment it acts, based on current human condition and prior expressed intent.
What began as a localized proof-of-concept has since matured into SOGA (State-Oriented Governance Architecture). SOGA transitions the project from a single demonstration into a protocol-neutral, execution-time governance runtime designed explicitly for complex, AI-mediated delegation chains..
View the original execution-time demo:
https://www.youtube.com/watch?v=u-uWl_s0PPM
The SOGA reference implementation and execution-time policy evaluation matrices are publicly available for technical review.
GitHub Repository: [Paste your GitHub link here]