Here’s what’s actually happening, stripped of the theatrical nonsense.

The race to build absurdly large closed models is indeed coming to an end. However, labeling this a “failure” or “crash” fundamentally misunderstands what has been happening in AI development.

What we’re witnessing isn’t a collapse: it’s a maturation. The field is moving from “how big can we make it?” to “how can we make it useful?” This is healthy. This is progress.

Let’s talk about the elephant in the server room. There’s indeed more to GenAI than models built with hundreds of billions of VC money, most of which has been efficiently transferred to Nvidia’s bank account and hyperscaler infrastructure bills.

But here’s what the doomsayers miss: this concentration of resources and capital, while creating obvious market distortions, has also democratized access to powerful AI capabilities. The same cloud platforms that charge enterprises premium prices also offer APIs that allow any developer to experiment with state-of-the-art open-weight models for a fraction of the cost.

The marketing of AI as the second coming of your deity or prophet of choice was always absurd. Every technology goes through this cycle:

We’re transitioning from phase 2 to phase 3, and the pundits are predictably losing their minds. But those of us actually building with AI? We’re already in phase 4, quietly shipping products that work.

You can absolutely get business value from GenAI today. The key is to ignore the noise and build from first principles. This means:

Think systems, not models. The model is just one component. Real value comes from how you integrate it into workflows, how you handle edge cases, how you measure success, and how you iterate based on real user feedback.

Measure ruthlessly. If your GenAI project doesn’t have clear metrics and ROI calculations, you’re doing it wrong. This isn’t research: it’s engineering.

Here’s the least sexy but most important truth: AI will continue to evolve exactly like every other field of technology. New architectures will emerge. Efficiency will improve. Costs will decrease. Capabilities will expand. Applications will proliferate.

Remember when “the cloud” was going to either revolutionize everything or was just “someone else’s computer” and doomed to fail? Now it’s just… where we run things. Remember when mobile apps were a bubble? Now they’re just… how we interact with services.

The Engineer’s Manifesto

Read. Not the headlines, but the papers, the documentation, the post-mortems, the actual technical content.

Experiment. Build things. Break things. Try different approaches. Test your assumptions. The cost of experimentation has never been lower.

Build. Ship actual products. Solve real problems. Create value. The best response to hype cycles is working code.

The Path Forward

Meanwhile, those of us who understand that this is just another technology — powerful, useful, but ultimately just a tool — will continue doing what engineers do: building things that work.

That is all.