Am I thinking about AI the right way?

Many of the more interesting companies I’m meeting these days are applying AI to some existing problem area. While every company and industry is different, I’m noticing a few themes and questions that seem to repeat themselves across companies. I wanted to share some of these here and actively invite your comments and suggestions.

Here are four of the questions I’m asking myself. Am I thinking about this right?

I. Sustaining or disruptive? To my mind this is a question about industries or verticals. In some verticals, I suspect that AI will be fundamentally a sustaining innovation in that it will strengthen the competitive position of incumbent players. This is often true where incumbents are sitting on tons of data or complex workflows and where challengers will have a very hard time breaking in. Generally, I think it’s easier for challengers to leverage AI to disrupt an industry when the table stakes for a useful product are lower. Killing Canva with AI might be pretty hard because Canva has a huge head start on features and community. Adding AI is not a heavy lift for them. By contrast, going after the container shipping industry with AI might be a bit easier.

II. Autopilot or copilot? A second question is whether the AI solution acts as a copilot (augmenting human capabilities by providing insights, suggestions, or tools for decision-making) or as an autopilot (fully automating tasks with minimal human intervention). There may be some instances where autopilots are high valued such as highly repetitive simple tasks or extremely complex tasks that humans struggle to perform. But in other cases, the copilot approach may actually create more long-term value. Copilot-style tools typically involve a complex human workflow, which may create stickiness. Copilot tools are also typically less likely to be perceived by buyers as mysterious black boxes, which might lead to faster sales cycles and easier customer penetration. Personally, I find myself drawn to both autopilots and copilots depending on the industry and the application. Which do you find more compelling? Why? And when?

III. Proprietary or shared? A third dimension is whether or not the AI solution is designed to handle proprietary data in a dedicated, often fine-tuned, customer-specific model, or if the AI solution benefits from sharing data across customers. This is usually application-specific. Some markets are characterized by proprietary data. This can complicate the sales process but can also lead to greater customer value and, perhaps, some barriers to entry as a company learns how to break down customer resistance. On the other hand, the promise of the network effects of insights gained across customers holds great allure. What are some good real-world examples of an AI company that has been able to leverage data across customers to achieve a competitive edge?

IV. Is the early-mover advantage linear or compounding? To me, this is the greatest unknown. Many companies (and investors) seem to be acting on the assumption that AI confers massive first-mover advantages that compound as a company scales. Companies selling AI for the legal industry, for example, seem to be betting that the more customers they have, the better their models will become, granting them an ever-increasing lead. This is a new, AI-specific, version of the famous “network effects” thesis that made USV so famous and successful. The argument does not necessarily rely on shared customer data, but only on the idea that proprietary models will get better as a company scales and encounters more samples. While I think this might be true, I am not certain. It’s possible that seeing large sample sets and encountering many customer use cases would begin to conder an insurmountable advantage on AI early movers in any given vertical. On the other hand, it’s also possible that open-source AI models are improving so quickly (and compute costs will come down so fast that these early mover advantages will melt away rather quickly). Could a late mover to the legal AI industry founded in 2029 displace Harvey or Leya with better models and cheaper computer? What do you think? Do the advantages of being an early mover in AI compound exponentially? Or will early movers be vulnerable to fast followers with better technology?

If you have thoughts on any of these four dimensions — or if you want to propose additional dimensions — please reach out!