Vibe coding is rewiring innovation. One can now quickly code applications using prompts and get usable code. This requires a lot of thought, but it is different from the thought that goes into writing code or doing UI design. These vibe coded applications and agents are also able to propose their own packaging and pricing. Today they do a bad job at this, even when the person doing the prompting knows a lot about packaging and pricing. But this will change. In the near future, will applications package and price themself?

AI advances through benchmarking. The major LLMs are systematically benchmarked as they are released and compete to appear at the top of the leaderboard. When developing FinAI Intercom fought hard to improve on its own benchmarking. Is there a general approach we can take to benchmarking AI agents for business? Will value models become the standard frame for benchmarking AI agents?