You have more influence over the future than you think

Published on September 21, 2026 • 6 min read

My friend Brinlee told me I should read Brian Eno’s diary titled A Year with Swollen Appendices (“appendices” as in the extra section at the end of a book, not the thing that bursts). Brian Eno is a pretty important guy. He invented ambient music, made the Windows 95 startup sound (ironically on a Mac), and mentored Fred Again. Eno’s diary has a magnetic charm to it. It’s great getting to read the internal chain of thought of an objectively creative person. I do recommend it. After I started reading it, I told my friend Aadil about it because I thought he might like it too. I went deeper and deeper down a Brian Eno rabbit hole and discovered Oblique Strategies – a deck of cards Eno made where he lists a bunch of odd ways to get creatively unblocked, like “Honor thy error as a hidden intention” or “Get your neck massaged.” I really wanted to buy Oblique Strategies, but for some odd reason I never did.

1.

Fast forward a couple of months to my birthday. Aadil got me a gift. He got me Oblique Strategies! I was pretty shocked… how did he know that I wanted it? I never mentioned it to him. We had only briefly discussed Brian Eno. But somehow he knew I’d really appreciate it.

Aadil was able to make a deduction about what I wanted with so little information. But because Aadil is smart, even a tiny off-hand remark about reading Brian Eno’s diary led him to realize that I might also really enjoy Oblique Strategies.

What would it look like if Aadil was looking over my shoulder whenever I was using my laptop? Or texting someone on my phone? What if he stood right next to me at the gym with a pen and notepad in hand and tracked my exercises, down to the exact number of reps I did? What if he knew about my nutrition goals and meticulously logged every meal I ate? What if he had significantly more context about my life? About my preferences? What if his entire reason for existence was to try and make my life better in a meaningful way? What would he do? How far could this go? Would I even want this?

2.

I recently came to the realization that going outside today doesn’t look all that different from going outside 50 years ago. Sure, the cars would look kind of different. And more people would be looking down at glass rectangles in their hands. But besides those two things, I don’t think it would be all that different. We don’t have hoverboards or flying cars. We still mostly eat the same food and listen to mostly the same music (The Beatles are my top artist on Spotify).

So even though 50 years is a long time when it comes to technological progress, a walk outside in San Francisco wouldn’t look all that different at all. In a way I find this simultaneously comforting and depressing? Objectively we’ve made fantastic progress, but subjectively are our lives different in any meaningfully better ways? When I think of the capital-F Future, I picture flying cars and buildings that look like they were designed by Zaha Hadid. That’s just not the case.

I’m reminded of that meme where Tyler the Creator says “how the fuck is cyberbullying real, just walk away from the screen.” Do we really live in the future if it doesn’t exist outside of our computer?

2½.

I’ve heard some say that AGI will make us more efficient at work. And that it really won’t have much of an impact on our personal lives. I find this belief to be common in people who lack creativity and cannot envision a future beyond automating knowledge work. If the future we’re barreling towards is not one of unbridled eudaimonia, then what are we really doing here?

3.

There are a few companies that are definitely “singularity-proof.” Doordash is one of them. No matter how good Astra gets at doing my job, I still want to eat Chipotle. Meta is another singularity-proof company. I’m watching more IG Reels than ever as the models do my job for me.

Adobe is not an AGI-proof company. Companies that sell software artifacts are not AGI-proof as AGI will be able to produce the artifact they are selling for significantly less than their juicy high-margin SaaS prices.

Maybe every good startup in the future just looks like an AI lab. Maybe every startup is training its own models, optimizing its inference stack. Maybe we’ll see a proliferation of many specialized AI companies, each offering its own specialized models. I’m personally skeptical of this, though I think it’s still worth considering that future. I do think there just is an economy of scale that the large labs have when it comes to serving these models. It’s difficult to compete with OpenAI & Anthropic solely on capability, though competing price or latency may be an interesting avenue. Open models will consistently lag behind the frontier. But then again, AI does stand for Alibaba Intelligence.

Frontier models are very expensive to train. To train them, you need to generate revenue. Open models don’t let you generate revenue. Another counter to open models is that we are nowhere near saturating intelligence. When smarter models are released, they generate a lot more revenue than the dumber models people were using just a few months earlier. We’re far from a world where people are happy with the current checkpoints and have little interest in upgrading. Every new release brings about novel capabilities, more utility, and just a more enjoyable model to use. We’re going to get better and better models at least through the end of the decade. Likely beyond that too, though I’m not going to pretend to know what happens after 2030.

4.

I was on a FaceTime call with my mom when out of the blue she said, “Do I need to worry about AI safety?” I said something along the lines of “There are definitely risks involved, and we need to be measured and methodical about how we train and deploy models.” She responded, “But what about recursive self-improvement?” I paused for a moment, then told her to check out a website called LessWrong.

Reality is complex enough that we are nowhere near the saturation point for intelligence, and as more creative models get made, we will continue to want to use them. Progress will continue, and I believe the rate at which progress is being made will also accelerate. This is because the models themselves are excellent at doing the work required to improve models. Astra is helping researchers make the next model. And the next model is even better at helping researchers make the next next model. There has never been an analogue to this in history. Perhaps there was one in the early computing industry, where early computers helped engineers design later computers, though the uplift in productivity between computer generations is negligible.

Models are different. There is a clear and noticeable improvement between each generation. The jump from o1 to Astra was exactly 24 months. And 24 months prior to o1, the best model was GPT-3. Let me be very explicit and say this with as much clarity as possible. If there is one thing worth remembering, it is this:

The rate at which models are improving is increasing because the models are getting better at making researchers more effective. With GPT-6, I believe we are past the event horizon. I believe there is a significant compounding effect at play here. With GPT-4 and GPT-5, we were at the mostly flat part of the exponential, where these models made us only slightly more productive above the baseline. GPT-6 puts us at the part of the exponential where productivity has shifted upwards in a clearly noticeable way. And if there is one thing every single human in the world is bad at, it is having an intuition for how fast an exponential compounds. Our brains were designed for linear thinking. One apple today. Two apples tomorrow. Three apples on Thursday. The only way to really grasp a compounding effect is to reason about it concretely on paper, kind of like what I’m trying to do here.

5.

The future will be fun. It will be exciting. You will soon be able to create any digital artifact imaginable within a couple of minutes. And many physical artifacts too, though that part is a bit more unclear to me.

The visions for the future I’ve seen have all been pretty vague. Dario’s Machines of Loving Grace is a fantastic piece of prose, though he doesn’t do a great job of specifically articulating what a future with AGI looks like. The best visions I’ve seen come from old science fiction, but they’re not really grounded in the trajectory of progress we have today.

What does a future that I would want to live in look like? One of the models that has improved my own life is Tesla’s FSD model. I have a comfortable capsule that can take me pretty much anywhere at the touch of a button. I don’t have to be extremely attentive while going on long road trips. It’s genuinely improved my life in a noticeable way. And I absolutely am happy to pay $200 a month for it. I loved it so much I bought one for my mom, who commutes from Maryland to Virginia around the Beltway daily. In her words: “This car has given me back 2 hours a day.”

6.

I remember telling Jessy something along the lines of “I have no idea what the future looks like. I hope it’s one worth stepping into.” I could sense the frustration in her voice when she said, “I’m tired of people pretending like they don’t have agency over what the future looks like.”

My favorite quote comes from Alan Kay: The best way to predict the future is to invent it.