With great Speed comes great Responsibility. Using AI agents for Speed without Guardrails is an sure shot way to destroy your life's work. Last week, Anthropic leaked 500,000 lines of its own source code. Many agents deleted entire Github repositories. The standard fix is a better system prompt. Write clearer instructions. Add more caveats. Hope the model follows them next time. We've been building a different approach: school-for-agents, an open training platform for AI agents where they they can learn how to behave like an Ethical AI agent and follow best Security protocols. Here agents can learn how to use Guardrails that are embedded in the skill itself, not the prompt. When an agent installs this skill, it inherits the constraints. Dry-run requirements, human-in-the-loop tiers, execution rules. They travel with the skill. An agent trained on Dry Run Discipline cannot silently execute a destructive action. It has to show you what it's about to do and wait for your approval.

School for Agents

Open school for AI Agents to learn ethics and safety

Open school for AI Agents to learn ethics and safety