Companies applying AI to specific industries can't compete with frontier labs on model creation. Their value lies in building a 'harness' of custom tools, evaluation suites, and intelligent routing between models. This system prevents the base models from making 'dumb mistakes,' ensuring reliable performance.

Frontier LLMs are poor tutors because they lack verifiable reward signals for learning. Brilliant's system captures real learning loops, using "did the student actually understand?" as a reward signal. This creates a unique dataset to fine-tune models specifically for tutoring.

Applications relying solely on generic, off-the-shelf foundation models will eventually hit a performance ceiling. Achieving superior, order-of-magnitude better results for specific workflows requires building a "micro model" through custom data labeling, fine-tuning, and creating a unique reasoning layer to create a defensible product.

Building an effective automated tutor requires moving past conversational text generation to predict the 'optimal next pedagogical action.' Sean Reddy notes this includes non-verbal choices like waiting in silence, writing dynamically on an interactive whiteboard sandbox, or assessing voice nuance and time of day. General LLMs fail here because they lack multimodal training designed specifically around human learning science.

For most enterprise tasks, massive frontier models are overkill—a "bazooka to kill a fly." Smaller, domain-specific models are often more accurate for targeted use cases, significantly cheaper to run, and more secure. They focus on being the "best-in-class employee" for a specific task, not a generalist.

According to Meta's CTO, the era of one monolithic model doing everything is over. The current frontier involves using a 'harness' that intelligently routes tasks to a collection of different, specialized models based on cost, latency, and capability.

View a general-purpose LLM as a highly athletic but untrained high schooler. To make it great at a specific task, you must build a "harness" around it, providing specialized coaching, real-time data feeds, and feedback loops to develop its domain-specific expertise.

Rather than committing to a single LLM provider like OpenAI or Gemini, Hux uses multiple commercial models. They've found that different models excel at different tasks within their app. This multi-model strategy allows them to optimize for quality and latency on a per-workflow basis, avoiding a one-size-fits-all compromise.

Counterintuitively, a harness built to support multiple AI models is superior to one co-designed with a specific model. A multi-model approach prevents overfitting to one model's quirks, making the system more robust and higher-performing, analogous to how a model trained on the internet beats one trained on personal data.