Siddharth Saoji Something interesting is happening in AI right now — and it’s worth paying attention to.
A small group of people are quietly shaping the direction of the entire industry. Whether it’s researchers from the big labs or influential practitioners like Andrej Karpathy, once they rally around a narrative, the whole ecosystem pivots to follow. The HTML pivot feels like a good example. And to be fair, this isn’t a bad thing — these are active, thoughtful builders, and in most cases they’re probably pointing us toward the best path forward.
But here’s the thing: when an entire wave is directed by a handful of voices, there are inevitably paths left unexplored. And those unexplored paths might be hiding real opportunities.
Is prompting still the best interaction for most things? We went from MCP to skills, from Markdown files to HTML, and the evolution keeps accelerating. What if we deliberately looked at the directions not currently in fashion and followed them to see where they lead?
One example I keep thinking about: open-source, edge-based models. Could they be optimized to do most of what the leading frontier models do today? Maybe, maybe not — I’m not a researcher, and I won’t pretend to have the depth the big labs do. But the question feels worth asking.
If the influential voices are always right, why does every cycle produce a few surprises?
What are we missing this time?