As AI models are used for critical decisions in finance and law, black-box empirical testing will become insufficient. Mechanistic interpretability, which analyzes model weights to understand reasoning, is a bet that society and regulators will require explainable AI, making it a crucial future technology.
Programming is not a linear, left-to-right task; developers constantly check bidirectional dependencies. Transformers' sequential reasoning is a poor match. Diffusion models, which can refine different parts of code simultaneously, offer a more natural and potentially superior architecture for coding tasks.
For consumer products like ChatGPT, models are already good enough for common queries. However, for complex enterprise tasks like coding, performance is far from solved. This gives model providers a durable path to sustained revenue growth through continued quality improvements aimed at professionals.
While AI labs could build competing enterprise apps, the required effort (sales teams, customizations) is massive. For a multi-billion dollar company, the resulting revenue is a rounding error, making it an illogical distraction from their core model-building business.
Glean spent years solving unsexy enterprise search problems before the AI boom. This deep, unglamorous work, often dismissed in the current narrative that credits AI for its success, became its key competitive advantage when the category became popular.
The enduring moat in the AI stack lies in what is hardest to replicate. Since building foundation models is significantly more difficult than building applications on top of them, the model layer is inherently more defensible and will naturally capture more value over time.
The process of struggling with and solving hard problems is what builds engineering skill. Constantly available AI assistants act like a "slot machine for answers," removing this productive struggle. This encourages "vibe coding" and may prevent engineers from developing deep problem-solving expertise.
Enterprise buyers purchase tools like Slack because employees love using them, not based on clear ROI. This presents a major adoption hurdle for non-viral, single-player products like enterprise search, which must find creative ways to generate widespread user adoption and love.
Anthropic's team of idealistic researchers represented a high-variance bet for investors. The same qualities that could have caused failure—a non-traditional, research-first approach—are precisely what enabled breakout innovations like Claude Code, which a conventional product team would never have conceived.