Jonathan Izcovich 🇫🇷/acc

AI Agent innovation is going to follow the same pattern of disruptive innovation that we've seen happen time and time again in technology. As most people in tech know by now, new disruptive innovation often starts out as tackling a set of problems for the lowest end of the market or an underserved part of the market with a new technology or business model, then advancing over time to tackle more and more of the market and use-cases as product performance continues to improve.  For instance, when the personal computer was first introduced, PCs were inferior to mainframes and minicomputers for high-end business tasks, but it introduced computing to the average consumer and knowledge worker; then over time, as performance improved and more software was introduced, PCs quickly took on more segments of the market. In digital streaming, the initial streaming products like Netflix often seemed inferior to cable providers, but over time, with better bandwidth in homes and more ubiquitous hardware support, streaming began to topple traditional television. We've seen this play out in many other technology categories, from digital photography to EV cars and cloud computing to smartphones. In each of the cases, the new technology starts out as being able to only serve a small fraction of the customer base that the incumbent could serve, then with continuous technological advancements, the disruptive innovation chips away at capability after capability to solve the entire need of the customer.   We're likely going to see the same dynamic likely play out with AI Agents. While we're in the earliest innings of AI Agent development right now, where most AI Agents can only handle a small fraction of work that we throw at them, and with often limited quality and accuracy. This is a dynamic that will inevitably change over time, and likely quite rapidly. The important thing about disruptive innovation is to understand what vector of improvement you can expect to continue to see over time that will cause the "low end disruption" to improve in quality over time.   With AI Agents, there are multiple components to a self-reinforcing flywheel that will serve to dramatically improve what AI Agents can accomplish in the near and long-term: 1. The underlying cost of GPUs will only continue to improve on a price/performance basis. 2. AI models are continuing to get more efficient over time, as we've seen with token pricing from GPT-3, 4, and others. 3. AI model quality is improving constantly, just as we've seen with Llama 3, GPT-4o, Gemini, and Claude releases recently.   4. The architecture of AI Agents will only continue to get more robust as time goes on, with more open frameworks. 5. The data and app connections that AI Agents can talk access with equally get more sophisticated. Altogether, these improvements mean that what we're seeing today in AI Agents is barely scratching the surface of what will soon be possible. Something that may not work today or seems too expensive to automate, may just be one performance improvement cycle away from becoming possible. It also means that architecting for the future is far more important than over-optimizing for just what is possible today.  When we first started Box at the beginning of the cloud wave, we had many competitors that were still building on-premises solutions because of the temporary limitations of cloud at that point. This turned out to be a very bad architectural decision because it eventually meant they were hobbled by a design choice that they were stuck with for years to come. The right move in a fast moving market that has many generations of improvements to come is to bet on the architectures continuing to improve rapidly and build for the future.