There have been two times in the recent history of AI development that I’ve texted my “normie” (non-tech) friend group to show them a new tool. The first was the introduction of Dall-e which gave us the ability to create images from text. It felt like a magic moment, and I was so excited to text the group asking for ideas to throw at the machine. The second time was yesterday, when OpenAI revealed their new SOTA image model. Yesterday we had the “Ghiblification of the Internet” where everyone on tech twitter was using the new tool to transform images of themselves into anime characters. It felt like magic again. For the past few years image models have made improvements, but nothing has left me excited enough to text the non-tech friend group. Even SORA felt like it wasn’t cool enough to show them.

So, I have a new metric. If I can’t text my non-tech friends about a new AI tool, it’s not big enough.

I think this is why coding agents have not felt as “transformative” as these image models. I’ve been using AI coding agents for over a year now, and their usefulness comes as more of a slow burn. When I started out with copilot it felt like a better auto-complete. Now, with Cursor, I ask the agent to write meaningfully large blocks of code. But there was a never really a moment where the model changed and it was OBVIOUS that we were in a new era. Looking back, it’s obvious that we are in a new era for coding. “Vibe coding” is the new craze, but it’s definitely felt like more of a slow burn to get here compared to the image models which slap you in the face with their capabilities.

In some ways the coding agent slow burn invalidates the group chat metric. I never had the “I NEED TO TEXT MY FRIENDS IMMEDIATELY ABOUT THIS” moment with Cursor. But it doesn’t mean cursor isn’t useful. Perhaps the group chat metric tells us when something is almost certainly going to be big, but the absence of reaching the group chat metric doesn’t mean failure.