Tyler Bettilyon

Tyler Bettilyon is a technology and education professional whose experience spans software engineering, computer science education and AI. His work is shaped by a practical approach to emerging technologies, with an emphasis on understanding how AI systems actually work, where they can deliver value and how professionals can continue learning as the field evolves.

From Teaching to AI Leadership

Bettilyon’s educational background shapes his approach to AI systems. After earning a computer science degree, he began his career as a full-stack engineer. A transition into education followed, with teaching roles in full-stack engineering and data science at a coding boot camp and later a small computer science school. He subsequently launched TEBS Lab. At Bradfield and TEBS Lab, AI became a major part of his teaching as neural networks advanced in areas such as computer vision, before the ChatGPT and large language model era.

The experience deepened his interest in AI while also shaping a critical perspective on how the technology is discussed. Bettilyon has written about and questioned the ‘magical’ and anthropomorphizing language used around large language models, noting that it can obscure how the systems actually work and create confusion about their capabilities.

The path to Cinnamon began years earlier when he taught the company’s CTO a computer architecture class. Later, the CTO approached him to deepen his understanding of AI and LLMs, leading Bettilyon to conduct a 10-course tutoring series. The relationship eventually led to an invitation to join Cinnamon.

Finding the Right Fit for AI

At Cinnamon, Bettilyon applies AI to automate the process of answering prior authorization questionnaires. After a physician prescribes medication, insurers may require proof of medical necessity through questions about contraindications, lab results or treatment history. Answering them can require searching medical records, making the process timeconsuming for healthcare professionals.

Cinnamon’s system integrates with customers’ EHRs to pull relevant records and provide the proposed answers along with the model’s reasoning and direct quotes from the EHR. Those quotes are validated against the original EHR text to give customers greater confidence in the results.

Bettilyon believes effective AI adoption starts with problems suited to what current models can reliably handle. Prior authorization fits because many questions involve clear semantic patterns. Diagnosis assistance presents a greater challenge as differential diagnosis requires narrowing numerous possible conditions. Deploying AI where it remains unreliable can quickly undermine user trust.

Preparing for AI’s Next Phase

Bettilyon expects LLMs to become more capable and reduce hallucinations, though he believes the rapid pace of raw-model development seen from 2020 through early 2026 is beginning to slow. He anticipates greater gains from agentic workflows, where models and other systems work together to overcome the limitations of standalone models.

Bettilyon also sees open-weight models becoming viable for many applications. Organizations can self-host or cloud-host them to save money and control models and data. As capabilities converge, a slightly less advanced model may soon perform many tasks well enough without the cost of a state-of-the-art model.

  ​Deploying AI where it remains unreliable can quickly undermine user trust.   

Cinnamon plans to expand into appeals and other prior authorization workflows beyond its current EPA process. These will require more from physicians, including reviewing policy documents, identifying requirements and determining what information belongs in letters to insurers. For Bettilyon, the path forward is iterative: establish what works, learn from it and expand one step at a time.

Leading Through Openness

His pragmatic view of AI is matched by an equally deliberate approach to leadership. Bettilyon’s leadership centers on openness. He encourages team members to challenge assumptions, flag AI and security concerns and speak up when they see something wrong.

His background in education also influences how he develops engineers. He emphasizes learning and growth, giving team members stretch goals and time to understand unfamiliar areas. Through one-on-one discussions, office hours and emerging-technology meetings, he encourages rigorous learning about what is new, what works and what does not.

He believes people should do the hard work of learning before relying heavily on AI. While LLMs can produce cogent answers, using them for homework or take-home tests can mean skipping skill development. He encourages professionals to use AI without letting their skills atrophy and to build an intuition for how the tools work. When approaching something new, he recommends making an initial attempt before using an LLM to identify and correct mistakes.