The Exponential
“The most surprising thing has been the lack of public recognition of how close we are to the end of the exponential.”
Dario Amodei, CEO, Anthropic
What is the “exponential”?
It’s not one curve, it’s two and the tension between them is the whole point.
As we can see from the blue curve in the above graph, AI has been improving exponentially for over two decades. Every year, model capability is dramatically increasing. Dario’s view is that compute, data quality, training duration and objective functions drive this and everything else is just noise.
The grey dashed line shows us that companies and teams adopting AI are moving linearly, slowly, carefully, constrained by security approvals, legal reviews etc,.
While AI capabilities move at a staggering exponential rate, the ability of companies to integrate them is limited by real-world constraints. These are two separate exponentials racing on different curves and mixing them up causes most AI prediction errors.
What the exponential actually changed for developers
The bottleneck has shifted. It used to be that AI wasn’t smart enough. Now it’s that infrastructure can’t keep up with how capable models have gotten.
1–2 years ago, for short focused tasks like debugging one test, lots of human steering was needed. Just a while ago, complex multi-step workflows were running overnight, resolving entire backlogs. Now, coordinated multi-agent systems are handling things like end-to-end M&A pipelines.
The model capability jump is real and fast, Claude went from Sonnet 3.7 scoring 62% on SWE-bench to Opus 4 scoring 87%, a 25% jump in about a year!
Anthropic went from planning for 10x growth to experiencing 80x annual growth in usage and revenue.
What should developers do about it?
Design for the future by building architectures that can absorb the next intelligence jump, not just today’s model.
Make upgrades easy by not building their systems tightly around how the current model behaves, that when a newer one comes out, everything breaks and you are back to square one. Keep it simple and flexible.
Give Claude more room for adaptive thinking and controlled tool access. Let it iterate on its own outputs instead of micromanaging every step.
All of this gives us a picture of how AI can transform the entire economy as the threshold for what a tiny team can accomplish has shifted dramatically.
But also, there is a catch! When you accelerate one part of the process, the human parts like security reviews, verification, design decisions suddenly become the new bottleneck. AI didn’t speed those up, at least not yet. Scaling isn’t just writing more code faster, it’s making sure everything holding that code up can keep pace too.
What “near the end” actually means
Referencing Amodei’s quote, this does not mean things are plateauing! Amodei uses “end” closer to the notion of “endgame”, meaning we are getting to the point where all benchmarks pegged to human ability will be saturated and AI systems will be better than any human at any cognitive task. The exponential isn’t slowing down, it’s simply approaching its destination.
Code with Claude 2026
At the event, Ami Vora, CPO at Anthropic, delivered what attendees called the strongest message of the keynote, even though model capabilities are improving exponentially, most organizations are still pursuing AI adoption on a linear path. There is a gap between what AI can do and what it is actually doing for people. Closing that gap is the developer’s job. “The Exponential” was the central theme of the entire Code with Claude 2026 keynote.
Every session at the conference circled the same shift, the gap between idea and shipped product is closing and the winners are the ones who planned for that instead of being surprised by it.
Vora spoke about how model capabilities are evolving exponentially, but corporate adoption is linear and the gap between those two curves is the problem Anthropic is trying to solve. The whole event was framed around that tension. Claude Code went from handling tasks measured in minutes in early 2025, to hours by end of 2025, to constant operation in 2026, that progression is what “the exponential” actually means in practice.
“The Exponential” is the WHY behind everything Anthropic shipped at Code with Claude. All the features they introduced are answers to “how do you keep up with capabilities that are growing faster than teams can adopt them?”
Source : Code with Claude 2026, San Francisco
