R
RESEARCH HUB
Blog
Future of Work AI×Tech March 2026 · 3 min read

Machines Are Accelerating.
Humans Are Buffering.

Intelligence is now on tap. The bottleneck isn't what AI can do. It's whether you can move fast enough to direct it.

The machines didn't slow down. They got cheaper, faster, and smarter — simultaneously. The people using them are still catching up to where things were six months ago.

That gap — between the speed of AI capability and the speed of human adaptation — is the defining tension of this moment. Not "will AI take my job?" That's the wrong frame. The right one: are you buffering or acting?

Intelligence Is On Tap

Andrej Karpathy said it plainly: "I had this intuitively wrong for decades. Agency is significantly more powerful and significantly more scarce than intelligence. Intelligence is on tap now so agency is even more important."

This matters more than it sounds. For most of human history, raw intelligence was the scarce resource. The smartest person in the room had leverage. Now intelligence — the ability to process, generate, analyze, synthesize — is abundant and cheap. Any serious thought work that used to take hours takes minutes. The cognitive ladder shortened overnight.

What's left? Agency. The ability to decide, direct, and move. To know what you want, point at it, and make it happen. That's the resource that didn't get commoditized.

The Context Gap

Here's the brutal part: agency requires bandwidth. You can only direct something if you understand it well enough to know where to point. And most people are still processing what AI could do in 2024, while the tools are already somewhere else.

Eleven tectonic shifts have happened in the last eighteen months. Inference costs collapsed 93%. Models crossed the threshold from assistant to agent. Reasoning became a purchasable feature. Entire job categories started contracting before anyone could write a policy. Knowledge that took years to accumulate started leaking out of experts and into prompts.

The machines ran eleven laps. Most people finished three.

The bottleneck isn't capability anymore. It's the human in the loop who hasn't caught up enough to give capable instructions.

Buffering Looks Productive

This is what makes it hard to see. Buffering doesn't look like paralysis. It looks like reading newsletters about AI. Watching demos. Attending talks. Staying "informed." It feels like preparation, but it's consumption — and consumption has no ceiling. You can consume forever without ever acting.

Agency is different. It's the moment you stop absorbing and start directing. You decide what you want built. You point at a problem. You iterate on output instead of just watching someone else's. The gap between consuming AI content and actually using AI to produce something is enormous — and most people are living on the wrong side of it.

The Speed Asymmetry

The machines will not wait for humans to catch up. That's not a threat — it's just physics. The compounding curve doesn't pause for comprehension. Capability doubles; the window to adopt at the frontier narrows.

The people who close the gap aren't the ones with the highest IQ or the best credentials. They're the ones with enough agency to act before they feel ready. To build something rough before they understand it fully. To direct before they're certain.

Machines are accelerating. That part isn't changing.

The question is whether you're still loading.

The 11 tectonic shifts of 2025–2026: inference cost collapse, model-to-agent crossing, reasoning commoditized, entry-level hiring stall, expert knowledge extraction, agentic coding, voice interfaces matured, multimodal default, frontier-lab consolidation, regulatory lag, and the consumer AI split. All happened while most people were still deciding whether to try ChatGPT.

Source: @ai_mind_set · Context Gap Annual Report · Feb 2026 · @karpathy on agency vs intelligence · Synthesis by Research Hub

Continue Reading

You Were Trained to Be Anonymous. 3 min → The End of Private Knowledge. 4 min →