Source: You Can’t Compete on Cheap Models Anymore
Type: YouTube Video Transcript
Author: Nate B Jones
Video ID: 1cSNE-ZkDLQ
Date: 2026-07-05
Summary
Nate Jones discusses the commoditization of cheap AI model execution and why the true value of AI has shifted to the “imagination layer.” When standard execution gets cheap, results converge and become generic because everyone is asking for the same things using the same playbooks. To differentiate and capture outsized value, individuals and companies must exercise technical-imagination—designing new workflows and asking frontier models questions that previously couldn’t even be formulated.
Key Takeaways
- Commoditization of Execution: Routing tasks to budget models (like glm-5-2 costing under $1) is a smart, cost-saving move that will soon be table stakes. However, because these models converge on the same standard tasks, cheap execution alone does not offer a competitive advantage.
- **The 40, reaching a performance level he could not have hit on his own. This task did not exist on any backlog—it was “scouted” and imagined by an expert.
- Redesigning the Building vs. Bolting on Motors: When factories first electrified, they merely bolted electric motors onto central steam-era driveshafts with minimal gain. Real productivity exploded when they redesigned the layout of the factories around cheap distributed motors. In AI, companies are currently “bolting motors” on by running old task lists through cheap models. Real transformation requires redesigning organizational layouts (e.g., team review cycles, validation systems).
- Stripe’s One-Day Migration: Stripe successfully migrated 50 million lines of code in a single day using AI. This was only possible because they spent years building rigorous task coverage, automated verification frameworks, and training people to drive models under high-complexity conditions.
- Manufacturing Imagination: Imagination is not an artistic gift; it is a technical and business capability born of hundreds of hours of hands-on interaction with frontier models (“fingertip awareness”). Companies cannot simply hire a single visionary to solve this because they lack the context. Instead, organizations must put domain experts in touch with frontier models and empower them with the permission and resources to make bets (e.g., posing $400 questions without approval).