What “Who Moved My Cheese?” Teaches Us About Foundational AI Models

If you haven't read Who Moved My Cheese?, it's a simple but powerful story about change, adaptation, and what happens when you refuse to move with the times.

The story follows four characters — two mice (Sniff & Scurry) and two tiny humans (Hem & Haw) — who live in a maze, happily feasting on their cheese. One day, the cheese is gone. The mice immediately move on to find more. The humans? They sit and wait, complaining that the cheese was unfairly taken from them.

Eventually, Haw wakes up, accepts reality, and finds new cheese. Hem stays behind, refusing to change — until he's left with nothing.

On the walls of the maze, Haw writes a series of lessons to remind himself (and anyone else willing to listen) how to deal with change. It's called “The Handwriting on the Wall”:

• Change Happens → They Keep Moving The Cheese
• Anticipate Change → Get Ready For The Cheese To Move
• Monitor Change → Smell The Cheese Often So You Know When It Is Getting Old
• Adapt To Change Quickly → The Quicker You Let Go Of Old Cheese, The Sooner You Can Enjoy New Cheese
• Change → Move With The Cheese
• Enjoy Change! → Savor The Adventure And Enjoy The Taste Of New Cheese!
• Be Ready To Change Quickly And Enjoy It Again & Again → They Keep Moving The Cheese

It's a lesson AI companies need to learn fast — because the cheese has already moved.

The Foundational AI Cheese Has Moved

For years, AI companies believed their models were their moat: our AI model is the most accurate! Our fine-tuned model is 5% better than the competition! Surely customers will pay for that! Spoiler: they won't.

Next, the logical move was to shift up the value chain. We'll build AI-powered meeting assistants! We'll do call center analytics! We'll build AI-powered voicebots! Unfortunately, so did everyone else.

Now? Tech giants keep making AI models cheaper, faster, and better. Workflow providers don't even need AI vendors anymore. Open-source AI is advancing so quickly that startups building foundational models must rethink their competitive edge.

The Need for Foundational Models Still Exists, But It's Harder Than Ever

Does this mean foundational AI is irrelevant? Not at all. There is still important research to be done in Speech AI and other foundational models, but niche players face an uphill battle. While general-purpose speech recognition is now largely commoditized, significant challenges remain:

• Live Speaker Diarization: Accurately distinguishing multiple speakers in real-time conversations, especially in noisy environments.
• Cross-Talk Handling: Overlapping speech is still a challenge for ASR models, especially in multi-speaker scenarios like call centers.
• Code-Switching: Many speakers naturally switch between languages mid-sentence, and current ASR models struggle to handle this effectively.

The Future of Foundational AI = Open Source + Proprietary Data

Some argue open-source models will dominate, especially after Google's leaked memo admitted open-source innovation is outpacing proprietary models. Does this mean proprietary models are dead? Not so fast.

Google I/O 2023 reminded us of an important truth: data still defines competitive advantage. Proprietary foundational models aren't just about better algorithms — they're about better, exclusive data:

• Real-time Data Access: Open-source models are powerful, but they can't access or process real-time insights.
• Proprietary Data Sets: Foundational models trained on exclusive first-party data offer unique advantages that open-source models can't replicate.
• Enterprise Security & Compliance: Proprietary models offer controlled environments, compliance, and reliability that enterprises demand.
• Personalization and Privacy: AI models trained on first-party user interactions can deliver deep personalization while maintaining privacy.

The Handwriting on the Wall for Foundational AI Models

• Change happens — proprietary AI still matters, but the game has changed. Value now lies in how models are deployed, scaled, and integrated — not just in how they're built.
• Anticipate change — the strongest AI companies won't be the ones clinging to proprietary models alone.
• Monitor the market — foundational models only retain value when backed by exclusive, real-time, and domain-specific data.
• Adapt quickly — those who solve the hardest speech AI challenges (like diarization, cross-talk, and code-switching) will outlast those competing solely on price.
• Move with the cheese — expertise, not just technology, is now the real differentiator.

Final Thought