Sarah Friar, the CFO of OpenAI, has shared what she calls a scorecard for the AI age. The point is not how much you spend, but how much value you get back. She names four measures: the amount of useful work done, the cost per successful task, dependability (meaning whether it works as expected), and the return on computing power. AI talk often focuses on huge spending. This framework tries to shift that attention back to results.

These four measures build a bridge between two groups. For managers on the ground, they offer a way to judge AI use with numbers, not just gut feeling. If a team lets AI handle customer questions, they can simply check whether the cost per case has fallen. For long-term investors, the same idea links a company's AI spending to its sales and profits. Flashy announcements matter less than whether the cost per result keeps going down.

The main battleground is shifting from the technology itself to securing talent. The Financial Times reported that Apple sent legal letters to dozens of OpenAI employees. The full intent has not been made public, but it shows the tension around the movement of AI talent. In the end, only a small group of people can build top models. Money can buy equipment, but know-how is much harder to buy. This suggests where the source of competitive advantage is moving.

At the same time, open source AI — models anyone can use — is gaining ground. Mozilla's survey gives an overview of where it stands today. Big closed models are powerful, but the fees add up the more you use them. Open models can run in-house, so they offer a way to rethink cost structure and vendor lock-in (being tied to one supplier). Here too, the scorecard matters. If the cost per result of open models nears a practical level, company decisions may change.

This trend is pushed forward by local AI becoming practical. LM Studio's new tool, Bionic, is an agent (an AI that carries out tasks on its own) that runs open models on your own PC. Its appeal is that you avoid pay-as-you-go cloud fees and keep your data in-house. It is not right for every use, but it feels realistic for teams that value cost and privacy. It shows that AI is no longer a matter of choosing only the big cloud providers.