Data (for agents) is the new oil

"Data is the new oil" — a phrase that defined a decade of tech. This idea fueled the "Big Data" wave, but over time, that hype faded. For most companies, simply collecting more data didn't unlock as much value as they'd hoped. More data wasn't always better.

But now, I think that relationship is changing.

In just the last two months, two major shifts have become clear:

  1. AI "agents" like Deep Research thrive on data and context. They are even more data-hungry than the large models that power them. As agents take on more complex tasks, they will require even richer data to perform well.

  2. These agents are no longer limited by reasoning ability. OpenAI's latest advancements—like o3 and inference-time scaling—are proof that LLMs are already surpassing human reasoning in many domains.

These shifts are leading to a step change in how data is valued and used. I expect two major consequences:

1. The Value of Data is About to Skyrocket

As AI agents become more capable, the owners of rare and valuable data will become more protective of it. We're still at the early stages—Deep Research is the very bottom rung of the AI agent ladder—but as we climb higher, proprietary datasets will become increasingly guarded. Companies will charge more for access. Some will build entire businesses around exclusive data ownership.

2. AI Agents Will Seek New Data

Because agents will be tasked with solving more complex problems, they will need access to data that doesn't currently exist in any dataset—public or private. To bridge this gap, AI will generate new data in three ways:

a) Self-synthesized data

AI models are already using reinforcement learning to improve their own capabilities in fields like math and programming. This trend will accelerate.

b) Human interviews

So much valuable knowledge exists only in human conversations—knowledge that has never been written down. This is why we built Ribbon API: to enable AI-human interviews at scale. The demand for expertise will explode far beyond today's expert networks (like Tegus or GLG), creating an entirely new economy where AI agents routinely consult humans for insights.

c) Sensor-driven data collection

AI agents will increasingly rely on real-world data from physical sensors. Some of these sensors will be personal (e.g., AR glasses), but many will be deployed at scale—collecting retail, weather, and industrial data to drive decision-making.

In the long run, I believe this will create an "agent-delegation economy." Just as Uber and DoorDash allow people to earn money by completing one-off transportation and delivery tasks, AI agents will delegate a wide range of real-world tasks to humans. Need in-person data collection? Physical intervention in a complex process? Human involvement will remain necessary in the AI-driven world, just in a different capacity.

We are at the start of something massive. The AI-agent revolution isn't just about better chatbots or more advanced assistants—it's about a fundamental reordering of how information is created, valued, and used.