Resources: guides and articles | Tokko Lab https://tokkolab.ai/resources/ Guides and articles from Tokko Lab on AI agents, how a team puts them into production, and the work that sits around that. en Score an AI agent on its own work https://tokkolab.ai/resources/ai-evals/ https://tokkolab.ai/resources/ai-evals/ Mon, 26 Oct 2026 12:00:00 GMT Take the cases from the process a person already runs, grade the state the system leaves behind, and run the same set again before every change. Guardrails before an AI agent runs https://tokkolab.ai/resources/ai-guardrails/ https://tokkolab.ai/resources/ai-guardrails/ Mon, 26 Oct 2026 12:00:00 GMT Write three limits before an agent can change a record: what it may read, what it may change, and what it must refuse. Then enforce them outside the model. Where a person stays on the path https://tokkolab.ai/resources/human-in-the-loop-ai/ https://tokkolab.ai/resources/human-in-the-loop-ai/ Mon, 26 Oct 2026 12:00:00 GMT A person takes the run in two situations: the retries are used up, or the next action cannot be undone. The approval only counts if the person can say no. How to build an AI agent that holds https://tokkolab.ai/resources/how-to-build-an-ai-agent/ https://tokkolab.ai/resources/how-to-build-an-ai-agent/ Fri, 23 Oct 2026 12:00:00 GMT Measure one process first, keep a fixed path wherever you can draw one, and let the model choose the next step only where that path cannot be written down.