Use Right-Sized AI Models to Cut Costs and Avoid Vendor Lock-In

At a recent tech conference, Notion's Sarah Sachs argued that most companies do not need the most powerful or expensive AI models for everyday tasks. She highlighted that simpler operations like summarization or file conversion can be handled adequately by smaller, less resource-intensive models. Sachs warned that locking into a single AI vendor or model puts businesses, shareholders, and customers at a disadvantage as AI pricing continues to rise. Her recommendation was to build AI systems in a modular way, allowing teams to switch vendors or models quickly as costs and capabilities shift. She also encouraged experimenting with open-weight models for mid-range tasks, emphasizing that 'good enough' performance often suffices and helps manage overall tokenomics.