The initial success of AI in coding is a natural outcome. Like early PC users who built tools for computers, software developers, as the primary early adopters of LLMs, logically focused on applying the new technology to their own workflows first.
Mobile networks built expensive global infrastructure with massive usage but captured little value as profits moved "up the stack" to apps. Foundation models, despite huge CapEx, face a similar risk of becoming a commoditized infrastructure layer with low pricing power.
AI's enterprise role is twofold. It will be embedded as a feature within systems like Salesforce to optimize specific tasks. Concurrently, it will operate as a top-level abstraction layer, pulling data from multiple systems (Salesforce, Workday, email) to generate novel, cross-functional insights.
AI will create a "consumer surplus" where productivity gains don't translate to higher margins. A task that took a week now takes a day, but instead of cutting costs, firms will simply do five times more analysis to stay competitive, passing the benefit to clients.
With past platforms like PCs or the internet, adoption could be predicted based on physical limits like hardware costs or broadband speed. With generative AI, a sudden algorithmic breakthrough could dramatically change price and capability overnight, making the future state far less predictable.
While consumer excitement for chatbots is high, the most tangible, high-demand use case that customers are "pulling out of your hands" is AI for software development. This has created a supply crunch and a narrowed focus in the tech industry from a broad 'everything' vision.
True opportunity in technology lies within uncertainty. Once a platform shift's winner is clear (e.g., Apple winning mobile), the strategic moment has passed. The most valuable focus for investors and founders is always on the areas where answers are still unknown and multiple paths are possible.
As AI moves into specialized fields like law or media, the critical questions become domain-specific, not technical. Like Netflix needing TV executives, the future of AI in these industries will be shaped by lawyers and producers who understand the nuanced problems, not just AI researchers in Silicon Valley.
Current e-commerce recommendation engines only understand SKUs and co-purchase data. AI can understand product attributes, style, and user intent on a semantic level, enabling previously impossible queries like 'suggest a coat that changes my look, but not too much.'
The most valuable startup ideas often identify latent problems that markets haven't articulated. This contradicts the idea that a generic AI tool can solve everything, as it requires a founder's unique vision to persuade customers that a previously unimagined problem exists and needs a new solution.
The enormous capital expenditure on AI by Google and Meta isn't just about positive ROI; it's a defensive, existential bet. They are driven by a fear of missing the next major computing platform and ending up irrelevant, like IBM in the 90s or Microsoft in the early mobile era.
Unlike operating systems which created strong developer and user network effects, foundation models lack a similar lock-in mechanism. Enterprises choose SaaS applications, which abstract away the underlying model, making the model layer a replaceable commodity rather than a defensible platform.
A key barrier for AI products is closing the gap between the 10% of daily active power users (often in tech) and the 40% of users who engage only weekly. This signals a product or UX gap, where mainstream users still see AI as a sporadic utility rather than an integral tool.