The Personalization Problem

The AI world has been captivated by the rise of the massive, all-knowing Foundational Model (FM). But for enterprises looking to leverage AI for specific business needs, the question is no longer "How powerful is the general model?" but "How personalized can I make it?"

This core tension, the shift from monolithic generalists to specialized, tailored solutions, was the focus of a dynamic discussion featuring Percy Liang, Co-founder of Together AI, and Wendy Gonzalez, CEO of Sama Group. Their insights highlight that successful enterprise AI is a journey of customization, driven by proprietary data and rigorous safety protocols.

The Smart Approach to AI Adoption

For companies just dipping their toes into AI, the best starting point is often the most accessible. As Percy Liang advised, enterprises should initially start with existing language models for prototyping. This approach is quick, efficient, and allows teams to validate their use case before investing heavily in customization.

However, once a business moves past the experimental phase and begins scaling its usage, the economics and performance demands change. The endgame is an orchestrated system where different models—some large, some small—work together, a departure from the simple idea of one model to rule them all.

The Data Edge: Customization and Evaluation

This is where the rubber meets the road. A foundational model knows everything on the internet, but nothing about your internal operations, customer interactions, or proprietary intellectual property.

Wendy Gonzalez highlighted her company's critical role in bridging this gap: fine-tuning models to an enterprise's unique domain. This requires moving beyond simply using off-the-shelf models and engaging in a sophisticated data process:

Tailored Data: Providing the specific, annotated data that allows the model to become an expert in the client's domain.

Comprehensive Evaluation: Gonzalez stressed that evaluation must go "beyond mere factual accuracy." Metrics must include factors like coherence and contextualization to ensure the model’s outputs are reliable and usable in a business setting.

The one thing that remains constant, even as models change, is the need for a robust and adaptable evaluation strategy.

Navigating Risk and Optimization

The conversation also addressed two critical, yet often underappreciated, factors in the modern AI landscape: security and efficiency.

Open Weight Caution

Both speakers acknowledged the evolving field of AI models and the necessity for adaptability in model selection, warning that enterprises must remain vigilant about what they deploy. They addressed the risks associated with open weight models—models where the parameters are available but the training data and code are not transparent—stressing the importance of caution due to potential security vulnerabilities.

The Shift to Inference

The current industry focus is shifting from the massive computational cost of pre-training models to the efficiency of inference (the actual process of running the model to generate a response). Percy Liang detailed Together AI's expertise in optimizing inference for large models, recognizing the huge cost savings and performance gains this provides for high-volume users.

Meanwhile, Wendy Gonzalez underscored a future where systems are not static. She highlighted the potential for future systems to improve through user interactions, suggesting a self-correcting AI loop.

This vision, paired with Gonzalez's strong emphasis on the need for responsible AI practices, confirms that building trust and ensuring ethical outcomes are paramount for sustained enterprise adoption.

Join the Conversation at HumanX 2026

The strategic application of AI—from model selection to data governance—is the definitive challenge for the modern enterprise.

We are thrilled to announce that Wendy Gonzalez will be returning to HumanX in 2026 to continue sharing her expertise on data curation, model evaluation, and building trust in AI systems.

Secure your spot soon to save time, money, and last-minute worry.