Two years ago, AI was a calculator. You asked, it answered. You prompted, it generated. Useful, but fundamentally passive.
Today, something shifted. The agents are no longer waiting for instructions.
The Math Problem
Daniel Litt made a bet in 2025: AI would have a 25% chance of writing a top-level math paper by 2030. He recently conceded. Not because the timeline accelerated — but because the capability gap collapsed overnight.
OpenAI and DeepMind achieved gold-medal level on the International Mathematical Olympiad. A project called "First Proof" gave AI 10 real research problems from working mathematicians. OpenAI solved 5 correctly. DeepMind's Aletheia got 6.
As Jeremy Avigad from Carnegie Mellon put it: "We have to face up to the fact that AI will soon be able to prove theorems better than we can."
The Economic Shift
But capability alone isn't the story. The economic model is breaking.
Greg Brockman laid it out plainly: "The inference compute available to you is increasingly going to drive overall software productivity."
In Silicon Valley, a fourth component is joining compensation packages: AI compute budgets. Engineers are negotiating token allocations. Some companies are adding "Copilot subscription" to benefits. Tomasz Tunguz estimates that 20%+ of engineering costs could come from AI inference usage.
The engineer who doesn't have compute access becomes the bottleneck. Not hours worked — tokens consumed.
The Hardware Wave
China is betting big. Wuxi city announced up to $720,000 subsidies for robotics and embodied AI projects built on agent platforms. Shenzhen's Longgang district offers $290,000 plus free office space for one-person startups.
The robots are coming. Affordable, programmable, vision-enabled hexapods like the miniHexa ($200) are making embodied AI accessible to hobbyists and students. The same tech that got a $720K subsidy in China is available on a credit card.
What This Means for Builders
We're at an inflection point. The tools exist. The economic incentives are aligning. The talent is waking up to what's possible.
At Tiny Little Lab, we're building accordingly. Agents that remember. Agents that reason. Agents that don't need hand-holding.
The question isn't whether agents become teammates. It's whether we'll be the ones building them — or the ones left behind.