The future of AI will be dominated by the companies that pick the right abstractions today.

The Internet today isn’t fundamentally that different from the Internet of the 90s. The hardware is faster and more reliable, but the core principals have not changed on a foundational level. TCP/IP is more or less the same. HTTP is basically the same. HTML has hardly changed at all. JavaScript has evolved greatly, but we still target old engines with cross-compilation and polyfills.

There was no groundbreaking tech that enabled the Internet to become so much more capable and pervasive today than it was in the past. The underlying hardware gradually improved, but many of the capabilities we have today could have been achieved much earlier with the right programming techniques. The biggest things that changed were the ideas powering it. It took us decades to refine and build out the abstractions, patterns, and frameworks behind the internet we use today.

Internet companies in the 90s and early 00s failed because they over-promised and under-delivered on a vision they presented but they could not deliver on. It was not because the underlying technology wasn’t capable enough. They chose the wrong abstractions, couldn’t deliver, got buried under tech debt, burned out, withered, and died. The same story is happening again today with AI, and just like with the Internet boom and bubble and bust, the winners and losers of the future will be decided by the abstractions we choose today.

We were told in that era to move fast and break things, and we still have to, but those that plan and think out their implementation strategies before even making their first moves will be far more successful. An upfront investment in planning out your system design is a small fraction of the investment of building things. If you build on the wrong abstraction you’ll build the wrong foundation and everything you’ve built on top of that will come crashing down.

Nobody knows what the right abstractions for AI are yet and most companies will be more focused on building things fast than building things right. We’re in uncharted territory where the most intuitive patterns for building reliable, scalable AI systems are still being discovered. Building fast without a plan is a risky gamble hoping you’ll stumble on the right patterns by accident. The companies that think through their abstractions early and choose them thoughtfully will be able to lead at the forefront of technology and refine their direction deliberately instead of scrambling to keep up.

If the progress of AI were frozen today and the foundational tech stopped getting better, there would still be so much untapped potential waiting to be unlocked by those who figure out the right abstractions, patterns and frameworks. How you manage data flow and context and sub-agents and tool calls are all pivotal to the capabilities, quality, reliability, and success of your AI agents. The abstractions needed to build with AI the right way are still being figured out on the fly.

Developers have started using AI to build things for them, but they’re just rehashing the same old patterns to do things we have already done. Over reliance on AI will lead you to recycling the existing abstractions that the AI was trained on. The breakthrough patterns and tools we need to build don’t exist yet but AI can help us get there faster if we use it to empower ourselves and learn and grow so we can make the ideas we need a reality. Those that use AI to be lazy and offload their work and thinking are only detaching themselves from developing the underlying abstractions needed to better utilize AI and they will fall behind.

The Internet bubble was disastrous for many, but for those that got it right, it was the greatest opportunity that could have been afforded to them. That same opportunity has emerged again with AI, and this time we can learn from the past to get it right. Those that can lead through the chaos will be the ones that succeed.