The AI revolution isn't about who has the most data. It's about who can take the most useful actions.
For the past decade, we've been told: data is the new oil. The companies with the biggest datasets would win. Collect everything. Store it forever. The moat is the data.
Something changed in 2024. And most people building products haven't caught up yet.
Today's AI Is Like a Super Student
Think of today's best AI models like a student who has read every book in every library. They have absorbed nearly all of recorded human knowledge. In a test, they outperform most human experts. In conversation, they are fluent in every domain.
But give them a task — a real task, something that requires a sequence of steps in the world — and the old models would just describe what they'd do. They could tell you the answer. They couldn't act on it.
The shift happening right now isn't about smarter students. It's about giving them hands.
What's New?
- Modern AI systems can now chain sequences of actions — not just predict the next word
- They can use tools: browse the web, write and run code, send emails, book meetings, call APIs
- They can plan multi-step tasks, track state, and recover from failures
- A single AI agent can now replace entire workflows, not just answer individual questions
This is the transition from AI as knowledge retrieval to AI as action execution. It changes everything about what a competitive advantage means.
Why This Matters
The old competitive moat was: we have more data than you. That moat is collapsing.
- Foundation models trained on nearly all of human knowledge erode proprietary data advantages
- Retrieval-augmented generation lets any model access any database at query time
- Synthetic data closes the gap for domains that were previously data-scarce
The new moat is something different: speed of action, quality of decisions, and depth of workflow integration. The companies that win won't be those with the most data. They'll be the ones who deploy AI agents that actually do things — fast, reliably, at scale.
What Should You Do?
- Stop collecting data for its own sake. Start building action pipelines.
- Ask: what decisions can an AI make better and faster than a human in my domain?
- Map your highest-cost repetitive workflows — automate the action, not just the analysis
- Experiment fast. The cost of AI experiments is now close to zero.
The question isn't "how do we prepare our data for AI?" It's "what can our AI actually do today — and what's stopping it from doing more?"
Final Thought
We're entering what Sutton and Barto call the Era of Experience — where AI learns by doing, not just by reading. Where the agent that acts accumulates advantage that no static dataset can match.
The question isn't "how much data do you have?" The question is: what can your AI actually do today?
Inspired by The Era of Experience (MIT Press).