Imagine a world where AI tutors are not just helpful but also trustworthy. This is the vision behind AI2T, a new system being developed at Carnegie Mellon University. Instead of relying on vast datasets, AI2T learns interactively, much like a human student. By demonstrating solutions and providing feedback, educators can teach AI2T to create intelligent tutoring systems (ITSs) for complex subjects. This approach allows for a more transparent and reliable learning experience than systems relying on large language models (LLMs), which can sometimes produce inaccurate or inconsistent information. AI2T's unique learning process involves building hierarchical task networks (HTNs), essentially step-by-step instructions that are refined through continuous interaction. One of the key innovations is STAND, a self-aware learning algorithm that helps AI2T understand its own learning progress. This allows the system to estimate its confidence in solving problems accurately and provides educators with a clear indication of when AI2T is ready to teach. This interactive approach not only saves time compared to traditional ITS development but also allows non-programmers to create sophisticated tutoring programs. Initial studies have shown promising results, with users successfully teaching AI2T complex concepts like multi-column addition and fraction arithmetic in under 30 minutes. While further research is needed to explore AI2T's full potential, it represents a significant step towards creating AI tutors that are both effective and trustworthy, paving the way for a more personalized and reliable learning experience for all.