The Pond Still Looks Empty

I thought I understood exponential growth.

I understand the math. Something doubles, then doubles again. Fine. But watching AI develop in real time is making me realize that understanding the math and actually having an intuition for it are two different things.

I wrote yesterday. Wasn't planning on writing again today. Then I listened to the ElevenLabs voice demo and started thinking about everything that's happened in just the past week.

Holy shit.

Imagine a pond with lily pads that double in coverage every day. It takes thirty days to cover the entire thing.

On day twenty-seven, it's only one-eighth covered. You could walk past, see mostly open water, and reasonably think, “Plenty of room.”

Day twenty-eight, a quarter.

Day twenty-nine, half.

Day thirty, full.

The last three days add seven times as much coverage as the first twenty-seven days combined.

That's the part I have trouble internalizing. My brain wants the next few days to look roughly like the last few days. A little more progress. Another update. Something I'll get around to eventually.

But each doubling starts from a bigger base. The amount being added keeps growing.

We've seen exponential trends before. Moore's law, for example. AI didn't invent that shape.

What feels different now is how directly these changes show up in things I can actually do.

Look at September. Anthropic introduced Fable 5.1 at the beginning of the month. Then on September 22, it released Opus 5.5, which Anthropic says performs at Fable 5.1's level on most work, while costing 40 percent less to run than the previous Opus. Today, Sonnet 5.5 arrived, promising faster, cheaper everyday work than the previous Sonnet.

OpenAI introduced GPT-6 Astra earlier this month, followed by Sol and Luna, bringing advances from that model into faster, more affordable options. Grok 4.7 arrived last week, too.

And on September 23, OpenAI updated ChatGPT Voice so you can use connected apps and plugins while you talk. Conversation becomes a way of operating software, creating things across your tools, all by voice.

That was five days ago.

And somehow I nearly left out Muse.

Meta released it on September 8, and then at Connect on September 23, it announced more connectors, voice features, and plans to bring it into AI glasses.

Meta's recap named ten retail brands being added, including Walmart, Best Buy, Sephora, and Wayfair, plus two payment services, Shop Pay and PayPal. That's twelve named retail and payment additions in the announcement, with some still forthcoming. The glasses integration is planned for the coming months.

The direction is clear. Agents that can talk, act, and plug into services people already use.

You don't need to memorize any of these names. The interesting part is how quickly impressive capabilities become more accessible.

A task becomes possible, then cheaper, then faster, then easier. Each change gives more people a reason to try it.

I'm especially excited about the creative side.

Imagine a simple, everyday thing. You have a family photo you love, but there's some random guy in the background.

Maybe you tried an AI editor a while ago and it removed the guy, sure, but it also changed your kid's face and turned a tree into something vaguely aquatic.

Fair enough. You decided it sucked.

Now imagine revisiting that same photo and getting an edit you're actually happy to share.

That's a meaningful change.

And then you realize you can animate the photo. Give everyone dialogue. Stitch a few scenes together. Make a short film. Start thinking about a sequel...

Anyway, I digress. I was just trying to remove some guy from a family photo, and now apparently I'm opening a film studio.

Today, I listened to the ElevenLabs v4 demo and almost laughed out loud. The emotional delivery was what got me.

Think about that alongside the September 23 update that lets ChatGPT Voice operate connected apps. You start to see where this could go: talking naturally with an assistant that can actually do things across your tools.

Emotion on one hand, utility on the other, and then whatever's coming next.

And then there's the rumor.

Whisper level. Take it or leave it.

Reports are circulating about an alleged Gemini 4 Pro preview that might beat Opus 5.5 in selected demos.

Maybe.

I'll wait for a real release and proper testing before handing out a trophy, but sure, Google, pull up a chair.

Here's the thing I keep coming back to.

A lot of people form an opinion, “I tried it, it wasn't ready,” and then they don't update that opinion.

I get it. Life is busy.

But when capabilities improve quickly, “not ready” can turn into “wait, what just happened?” faster than we expect. The pond is an illustration of compounding, not a claim that AI follows one exact doubling schedule.

The good news is the surprise isn't only on the scary side. Things that frustrated you last month might actually delight you this month.

We're all still learning how to live with that pace of change. I'm writing this because I'm trying to learn out loud.

The pond still looks empty.

Until one morning, it doesn't.