Robot Ventures Partner: In the Era of AI Agents, the Turing Test is Outdated

Core Viewpoint

ChainCatcher Selection

2026-01-21 15:18:21

Collection

We should ask: Can the machine trade with Brad and win all his money?

Author: Maddie P, Partner at Robot Ventures

Compiled by: Hu Tao, ChainCatcher

Stop looking at papers. The cutting-edge field has evolved to a new stage, while you are still refreshing arXiv like you refresh your ex's Instagram.

I know, because I did the same thing. For six whole months, I was obsessed with various model releases. Every benchmark was saved. The GPT-4.5 preview? I read it all. Do you know what I learned? Nothing. Do you know what I got? Less than nothing. I paid the price for my mistakes.

Meanwhile, a kid made more money with a model from three months ago and a Stripe API key than my "cutting-edge implementations" combined… even more than all my work put together. Not because his AI was smarter, but because his AI could directly access money.

Every AI researcher I know is now facing the same nightmare: they spent five years building God, only to find out that God needs a credit card and permission from Apple to exist.

The model wars are over. Not because someone won, but because winning and losing no longer matter.

The speed of intelligence commoditization is faster than that of permission commoditization.

In 2023, the price of GPT-4 was $30 per million tokens. Now, Gemini Flash Lite costs only $0.08. In just 18 months, the price has plummeted from luxury levels to cheaper than a morning coffee, a complete collapse.

Performance is converging. Pick any benchmark: ARC, MMLU-style exams, GPQA-style hard tests, code evaluations, the results are the same. Cutting-edge technology is advancing, but the gap between top models is narrowing relative to real-world production environments. It looks like the smartphone market in 2015. All products use the same chips, the same screens, the same cameras. The only real difference is who can get the carrier's authorization.

Models will continue to get smarter, but the speed of intelligence commoditization far exceeds the speed of permission opening. The gap between what AI can do and what it is allowed to do is growing, and therein lies immense profit.

The Corpses

The graveyard is filled with companies that had perfect AI but no authorization.

I witnessed a transcription startup go under, with accuracy even surpassing human stenographers. The reason for the collapse was not technical failure, but the inability to convert users into revenue. The product was great, but without payment channels, it was goodbye.

Another team developed a truly profitable AI trader, but was rejected by all app stores. Autonomous financial activities made compliance teams uneasy. There are currently no clear approval options for AI that can transfer funds without human oversight.

This is the real dilemma that will make you suffer. Even if customers are willing to pay, they will be frugal. Since they can wait six months for the next generation product to offer the same features for free, why buy a supercomputer? This is just like Intel's usual tricks in the 1980s. Your customers are not just hindered by permission restrictions; they are also waiting for you to exit the market. Commoditization is eroding your pricing power from the bottom up, while permission layers are squeezing you from the top down.

These two pressures compound. You can't charge enough to survive because next quarter's model will be cheaper. You can't quickly integrate payment systems because the pace of compliance processes depends on human speed. You are being squeezed from both directions. The ultimate winners are those who own the railroads. Regardless of how much you charge, whether you ship this quarter or next, they can collect tolls. They always get paid.

Companies do not die from bad AI, but from being trapped with excellent AI. On one side are permission restrictions, and on the other is commoditization pressure. This double bind is the strategy that constrains business development.

The Permission Stack

The value creation of modern AI lies in the accumulation of permissions. This is not because the models themselves are not good enough, but because autonomy is restricted.

Think about self-driving cars. We have impressive demos, but most systems are still just driving assistance. Humans are still involved, responsibility is outsourced, and the real world is full of extreme situations. AI is at the same stage. We have reached Level 2 autonomy in cognition, but have not yet achieved Level 5 autonomy in economics.

Level 1 Permission (Access): Can you call the API? → OpenAI = Thinking permission;
Level 2 Permission (Compliance): Can you store user data? → AWS = Memory permission;
Level 3 Permission (Revenue): Can you process payments? → Stripe = Charging users permission;
Level 4 Permission (Distribution): Can you reach users? → App Store = Publishing permission;
Level 5 Permission (Capital): Can the system obtain credit, margin, and settlement guarantees? = Economic activity permission.

Most AI companies are stuck at the third or fourth level. They can sell subscription services but cannot achieve financial self-sufficiency. This is why user conversion rates plummet. Each additional layer of permission is like a toll booth, and toll booths accumulate. Even if you have an excellent model, you still cannot achieve economic autonomy.

This is the real key: what we lack is not intelligence, but economic autonomy. Unless entities can borrow, settle, and repay under established constraints, the flourishing development of AI will only evolve into a transfer of wealth from application developers to the authorization layers that grant them the right to exist.

Why Agents Are More Expensive Than Prisoners

Here’s an experiment: your agent has an IQ of 180, can access a browser and command line interface, and can call any API. But it has no money and cannot borrow money. What can it build?

Nothing.

It's like a talented employee who can't even get reimbursed for a domain name; like a trader who can draft an entire business plan but can't pay for a $12 server hosting; like a trader who discovers the perfect arbitrage opportunity but can't execute the trade; like an F1 car in a world without roads.

Why not just give it a company credit card?

The premise of a credit card is that you exist as a person. You need a social security number, a bank account, and a three-year credit history. Your agent has none of that. It only has an API key.

Even if you could give it a card, it would be like giving an F1 driver a bicycle. Sure, it can move, but its utility is wrong. Credit cards limit your ability to take risks. They have daily limits, fraud alerts, and manual reviews. A broker might need a million dollars to complete a perfect arbitrage in thirty seconds. Good luck explaining that to Visa's fraud department. Moreover, credit cards are designed around predictable risks. They assume your spending follows certain patterns: groceries, gas, the occasional vacation. Underwriters can model this distribution. Brokers operate differently. They need to take on risks that cannot be predicted in advance. The point of intelligent systems is to discover unexpected opportunities. You cannot pre-approve credit limits for business opportunities that have yet to be discovered.

But the real problem goes beyond that. Credit cards are for employees to buy office supplies, while brokers are not employees. They need to borrow against future income, leverage locked positions, and flexibly adjust credit scales based on market opportunities. They need funding that can keep up with the speed of their thinking.

We have given agents various tools: browsers, terminals, APIs. Does it feel like victory is in sight? But in reality, any significant economic activity requires capital; and the infrastructure does not exist because the financial system was originally designed for humans. Business hours, various forms, someone always has to take the blame, and compliance officer Brad is at lunch from noon to 2 PM.

Why Crypto Is the Answer

Agents need funding that can operate at machine speed, without human approval, and can take on unpredictable risks. Where does this infrastructure currently exist?

On-chain.

This is not about the hype of Crypto, but the architectural reality. Traditional finance has human factors built into it from the ground up. This is not a flaw, but the essence of its business model. Every approval, every review, every compliance check means someone's job, someone's salary, someone's lunch break. This friction is the product itself.

Crypto eliminates human factors. This is not out of ideology, but due to technical limitations. Smart contracts either execute or they don't. No regulatory personnel need to check, and no fraud department needs to be contacted. The code runs the same way at 3 AM and 3 PM.

More importantly, on-chain collateral solves the underwriting problem. You do not need to rely on an individual's credit history or employment verification for lending; you rely on mathematics. Collateral is verifiable, liquid, and does not require trust. An agent holding staked ETH does not need to prove its value to Brad. ETH is locked and can be accessed programmatically. The issuance, use, and repayment of loans can all be completed in a single atomic transaction.

This is the crux of the matter. It is not because Crypto is cool (though it is indeed cool), but because cryptocurrency is the only infrastructure that allows traders to take financial risks without prior permission. Traders can borrow against locked positions, execute trades in milliseconds, and settle without intermediaries. The infrastructure already exists; it just has not been optimized for this application scenario… until the advent of Sprinter.

Sprinter (What We Are Building)

Last year, I tried to borrow against staked Ethereum (ETH) and told Michael Cieri about it. Three weeks later, I had my fourth call with Brad from Traditional Finance Inc.

Brad wanted to know my investment goals. Brad needed my proof of employment. Brad was concerned about the volatility of cryptocurrency. Brad went to lunch.

My Ethereum was just sitting there. Immutable. Verifiable. Continuously generating yield. And I was listening to soothing jazz while Brad was "confirming with his supervisor."

That’s when I realized: the entire credit system was orchestrated by Brad. Hundreds of Brads passing documents back and forth, charging for Brad's time.

Sprinter removes Brad.

Sprinter is a programmable credit engine. You can borrow consumable stablecoins using on-chain verifiable collateral without selling the collateral. Credit limits are restricted: funds do not go directly into a free wallet but can only be used through authorized routing, and repayments are prioritized. We first launched a consumer credit card as a distribution channel, then released an SDK so applications and agents can apply for short-term credit limits under strict constraints. Credit limits are provided in the form of an API. A rules-based underwriting mechanism. No Brad involvement required.

Not limited to humans. Humans are just the beginning. We are also building systems for machines that need to borrow at 3 AM to rent computing resources before prices spike. We are also building systems for agents that need to borrow against locked positions to execute 30-second arbitrage. We are also building systems for protocols that need loans in milliseconds rather than business days.

Other companies are developing products for humans using AI. We are developing products for both humans and AI using capital.

This difference may sound subtle, but its impact will reshape the economic landscape.

What Truly Disrupts?

Policy is the kill switch. A regulatory letter is enough to destroy everything. I have seen three perfect teams collapse within 48 hours simply because someone in Washington discovered their use case.

The settlement process becomes a political issue. You will see two economic systems: one is compliant dollars slowly circulating in banks, and the other is programmable dollars flowing at the speed of light. The gap between the two is not coincidental; it is a complete industry.

Stablecoin redemptions are like bank runs with a better user experience. The balance of the entire AI economy is built on stablecoins remaining stable. Once (not if) stablecoins experience volatility, the entire economic system will be affected. Of course, we might as well pretend that the risk is "blockchain throughput."

The Endgame

Someone will build a billion-dollar company using a model from three years ago. This is not because those models are particularly good, but because they found a way to use unfiltered capital for thoughtful, wise risk-taking.

Stablecoins decouple from decadence. Trading volume shifts from derivatives to actual transactions. Traders pay for computation, data, and reasoning.

A major power panics. "Unauthorized autonomous economic activity." This will make headlines. The market crashes. I’ll grab some popcorn and watch the show.

Company valuations are often based on API access rather than the quality of AI. Banking relationships are more valuable than model parameters.

The "two-dollar" economy era has arrived. Traditional dollars circulate in banks at human speed, while programmable dollars settle on-chain instantly. The arbitrage between the two becomes the real game.

If none of this happens, then all my previous judgments were wrong.

But I am not wrong. The evidence is here. The cost of reasoning has dropped by 99%. All models are converging on the same benchmarks. Companies that had perfect AI but no authorization have gone under. Brad still has lunch from noon to 2 PM every day.

What If I Am Wrong?

Perhaps GPT-7 is incredibly intelligent, and permission issues are irrelevant. Perhaps Brad will learn to use email.

But perhaps not.

The speed of intelligence commoditization far exceeds the speed of permission proliferation. Distribution channels remain limited. Permissions are still precious.

The AI investment game that everyone was playing is over. Not because the game failed, but because it was so successful that it became worthless. Investing in the smartest models, building the best RAG (Red, Amber, Green) models, optimizing reasoning—none of that matters.

The new rules of the game are permission accumulation, control surfaces, and economic tracks.

Winners will not have the best AI but will have AI that can do things with known information.

This is not pessimism. The development patterns of the internet, mobile, and cloud computing are all the same. Technology commoditizes, while railroad systems create value.