What is a falsifiable prediction?

A falsifiable prediction is a claim with specific measurable criteria and a timeframe, making it possible to definitively prove it true or false. Karl Popper established that a claim is scientifically meaningful only if it can be proven wrong. Every prediction in this series names a date, a measurable threshold, and a falsification trigger — the exact condition under which the author will admit the prediction was wrong.

Why 15 predictions and not a single headline claim?

Philip Tetlock's superforecasting research at the University of Pennsylvania showed that prediction accuracy improves dramatically when forecasters assign specific probabilities, track predictions over time, break complex claims into sub-components, and acknowledge uncertainty explicitly. Tetlock's trained superforecasters outperformed CIA analysts with access to classified intelligence by approximately 30 percent — the difference was methodology, not information. Fifteen specific predictions covering infrastructure, economics, technology, markets, and society produce a structured scorecard. A single headline claim cannot be verified; fifteen specific claims can.

When can these predictions be verified?

Four predictions are verifiable by 2028 (PRED-002, 005, 007, 013). Six more by 2029 (PRED-001, 004, 006, 008, 012, 015). Four by 2030 (PRED-003, 009, 011, 014). The final prediction, PRED-010, reaches its verification date in 2031. Every prediction will be revisited annually in a public verification post.

What happens if a prediction is proven wrong?

The annual verification commitment ensures the author says so publicly. Every prediction has an explicit falsification trigger — a specific condition under which it is wrong. If that condition is met, the verification post will say so, explain why, and say what it changes about how to read the rest of the scorecard. Willingness to be specifically, publicly wrong is the rarest quality in technology prediction.

Where do these predictions come from?

The 15 predictions are the content of Chapter 9 of The AI Agent Economy — Book 1 of an ongoing series by Atin Agarwal. Each prediction is derived from frameworks built across the preceding eight chapters: the dependency layer thesis, agent economics, the trust problem, India's structural advantages, the vibe coding quality gap, and the dharma framework.

From Book 1 · Chapter 9

15 Falsifiable Predictions About the AI Agent Economy

"A prediction that cannot be proven wrong has no intellectual value."

Technology prediction is a graveyard of vague claims. "AI will transform healthcare." "The metaverse will change how we work." These are not predictions — they are truisms dressed in future tense. They cannot be proven wrong because they never committed to anything specific enough to be wrong about.

These fifteen predictions are different. Each one names a date. Each one defines measurable criteria. Each one carries a confidence level. And each one includes a falsification trigger — the specific condition under which the prediction is proven wrong. This is Karl Popper's standard applied to the AI agent economy: a claim is meaningful only if it can be falsified.

Ten of these fifteen predictions can be checked within three years. Every one is reviewed annually in a public verification post. If any of them are wrong, the review will say so — with the same specificity as the original claim. Willingness to be specifically, publicly wrong is the rarest quality in technology prediction. That is the bet this page makes.

See glossary: falsifiable prediction · prediction scorecard

predictions with falsification triggers

2028–2031

verification window

10 of 15

checkable within 3 years

Annual

public verification reviews

Infrastructure Predictions

What gets built. The internet needed infrastructure before it needed applications. The agent economy will follow the same pattern.

PRED-003 By December 2030 · Confidence: Informed bet

The agent dependency layer will be a $100 billion-plus market by 2030.

Combined agent infrastructure revenue exceeds $100B/yr and surpasses the agent application layer.

15 Falsifiable Predictions About the AI Agent Economy

Infrastructure Predictions

The agent dependency layer will be a $100 billion-plus market by 2030.

A dominant agent orchestration platform will emerge with over 30% market share by 2029.

Cryptographic attestation of agent actions will be mandatory for enterprise deployment by 2030.

Economic Predictions

At least 10,000 solo operators will run five or more revenue-generating businesses simultaneously by 2028.

The median cost to deliver a SaaS feature will drop 10x for agent-first companies by 2028.

At least three major SaaS categories will see 50%+ price compression by 2029.

Technology Predictions

AI agents will execute more business tasks than human employees in software companies by 2029.

At least five new agent-native job categories will exist by 2028.

AI-generated code will cause at least three major security breaches by 2028, each exceeding $100 million in damages.

Market Predictions

A trust layer company will become more valuable than any single agent application company by 2031.

India will rank number two globally in agent infrastructure company revenue by 2030.

India will produce more agent-first SaaS unicorns than any country except the United States by 2029.

Societal Predictions

20% of top-tier VC-funded startups will have fewer than five employees at Series A by 2029.

Agent oversight will become the fastest-growing white-collar job category by 2030, while traditional knowledge-work hiring declines 25%.

At least three major economies will enact agent accountability legislation by 2029.

The AI Agent Economy — full analysis

Get annual verification reviews before they're published