Thriving Through Platform Shifts
Insights on Navigating the Rapidly Changing AI Stack
The AI ecosystem is shifting faster than ever. In just a few months, we’ve seen the arrival of Llama 3.2 and OpenAI’s latest models, with each release rewriting the rules of the game. For founders, CTOs, and product leaders, the real challenge isn’t simply keeping up—it’s deciding which infrastructure bets to make, and which abstractions will actuallyhelp you deliver value, knowing that your stack could be obsolete in a matter of months.
This kind of technological upheaval isn’t new. It mirrors what I experienced during the early days at Capillary, where we had to navigate a similar shift in the cloud landscape. While platform shifts can be daunting, they can also act as powerful tailwinds for those who get it right. The challenge lies in making the right decisions under conditions of uncertainty. Drawing on my experiences at Capillary and working with AI and ML systems, here are six key principles to help you steer through rapid technological evolution.
1. Velocity Trumps Perfection
When the landscape is evolving this fast, speed becomes your strategic edge. Don’t get bogged down trying to make the “perfect” infrastructure choices. In six months, the underlying technology may shift dramatically, making today’s perfect solution tomorrow’s bottleneck. Instead, focus on delivering customer value quickly and iterating based on feedback. You’ll likely need to rewrite parts of your stack in a couple of years anyway, but by then, you’ll have far better insights into your customers’ needs and product direction.
Personal Example: Back when we were building Capillary, cloud technology was still emerging. EBS didn’t even exist at the time. New features and services were announced almost weekly. Instead of trying to build an airtight future-proof architecture, we focused on delivering value immediately. That meant we could quickly adapt when AI, ML, and BI analytics began driving significant business outcomes. Years later, with more clarity, we made better-informed architecture decisions.
2. Focus on Invariants
Even in the midst of rapid change, certain fundamentals remain stable. These are the elements you can count on to guide your product decisions—think core customer needs, compliance requirements, and the operational value your product delivers. While the technology stack may shift, these constants act as a compass, allowing you to maintain focus on what truly matters.
Personal Example: At Capillary, integrating with external POS systems was anything but glamorous. It wasn’t the sexiest part of the tech stack, but it became a key differentiator for us over the long term. These “boring” aspects of your product often end up being the backbone of your business, so don’t overlook them while chasing after the next shiny thing.
3. Keep It Simple
In times of rapid innovation, it’s tempting to dive into every new framework or abstraction layer—whether it’s Langchain, AWS SageMaker, or Llama models. But layering too much complexity on top of an already evolving technology stack can quickly backfire. Simplicity should be your default. Use lightweight constructs until you have a clearer picture of your needs, then introduce complexity as it becomes necessary.
Example: Many companies are eager to integrate OpenAI APIs directly or through various frameworks, but these APIs are changing so fast that the added abstraction can often slow down progress instead of speeding it up. Abstractions are only valuable if they make things simpler or more flexible. Otherwise, they can become just another burden.
4. Invest in Quality Infrastructure
While your technology stack will undoubtedly change, the perceived quality of your product should remain stable. Your customers won’t care about your internal stack evolution—they’ll care that your product is reliable, performs well, and delivers value. To ensure this, invest in solid infrastructure that allows you to rapidly test, migrate, and compare across versions. Quality infrastructure enables you to maintain high velocity even through inevitable transitions.
Personal Example: Migrations are notoriously painful. Every time we’ve had to undertake one at Capillary, it required a massive effort. But having the right tools—and a team experienced in making these transitions—made a significant difference. In today’s AI world, investing in a robust evaluation (Eval) suite to compare models and infrastructure is essential.
5. Balance Engineering Productivity with Cost Awareness
Early-stage companies often emphasize speed over cost, and that’s usually the right call. However, if left unchecked, infrastructure costs can spiral out of control. As your product begins to scale, you need to keep an eye on these costs, especially when runway becomes a concern. Always have a baseline understanding of where your dollars are going, even if cost optimization isn’t the immediate focus.
Personal Example: At Taro, a pivot forced us to drastically cut our infrastructure costs. Because we had been tracking costs and had a clear view of where we were spending, we were able to quickly reduce them and extend our runway. If we hadn’t had that visibility, we likely wouldn’t have survived the transition.
6. Stay Close to the Cutting Edge
Even if you aren’t constantly evolving your systems, it’s crucial to stay informed about the latest developments in the tech ecosystem. Encourage your team to experiment, host internal knowledge-sharing sessions, and make sure they are keeping up with the rapidly changing landscape. Building a point of view on where the puck is going will help you make informed, strategic investments in the areas that will matter most as your product scales.
Personal Example: At Capillary, we initially built our analytics on traditional databases. But we stayed close to the emerging technologies of the time and quickly pivoted to a Hadoop-based system when it became clear that it would give us a competitive edge. This agility gave us a significant advantage over slower-moving incumbents.
There are always exceptions: Cohesity is a great counter example for all of this (and was built towards the lagging end of the hyperconvergence platform shift). Mohit had already written a distributed file system twice in his past life (Google file system, Nutanix) so his third file system architecture was near perfection. My sense is that the company still relies on the code that he wrote a decade ago! Hard to get this level of precision unless you’ve built the same thing before.
The rapid evolution of the AI stack isn’t a temporary phenomenon—it’s the new normal. As CTOs and founders, we must stay adaptable, focusing on velocity and simplicity while remaining grounded in the unchanging fundamentals of our business. Navigating these shifts effectively can make the difference between leading the charge or being left behind. I’d love to hear about your experiences navigating similar upheavals. What strategies are you using to stay ahead in a world of rapid change?