No main AI firm has endorsed crypto buying and selling bots. No frontier lab is coaching fashions for it. But a rising variety of merchants are utilizing Anthropic’s Claude to construct automated Polymarket bots, claiming earnings within the hundreds of thousands. Viral threads recommend anybody can do it.
However the loudest winners are utilizing methods any quant fund might replicate in a single day.
Three Assumptions, Zero Ensures
The narrative rests on three assumptions. Massive tech will ultimately construct purpose-made buying and selling fashions. Particular person merchants can maintain an edge in opposition to establishments. Autonomous AI brokers can reliably earn cash in open markets.
Haseeb Qureshi, managing associate at Dragonfly Capital, disagrees on all three counts. Within the Bankless interview, he pointed to legal responsibility threat, market construction, and the commoditized nature of AI. Collectively, these forces make this gold rush far much less promising than it seems.
The Legal responsibility Lure
Qureshi says that constructing AI for blockchain duties is technically trivial. An EVM simulator can take a look at looped lending or token swaps with ease. The fashions are succesful. They only haven’t been pointed at crypto.
The reason being institutional, not technical. First, crypto carries reputational baggage that AI labs need no a part of. “Crypto’s form of cringe,” Qureshi stated.
However the true barrier is legal responsibility. Think about Claude botches a leveraged commerce and wipes out $2 million. Or sends $10,000 to a burner tackle by mistake. No disclaimer can be loud sufficient to stop the backlash.
“It should 100% occur,” Qureshi stated. “Anyone who has a nasty expertise, it’s going to go tremendous viral.”
He in contrast managing a consumer’s crypto pockets to injecting unregulated Chinese language peptides. The draw back dwarfs any income upside. Coding recommendation gone incorrect is embarrassing. A drained pockets is a lawsuit.
Anthropic has already printed analysis on AI and blockchain. Its SCONE-bench research examined how nicely frontier fashions exploit sensible contract vulnerabilities. However that is cybersecurity analysis, not a product roadmap.
The inflection level will come from competitors. When one lab decides crypto quantity is simply too strategic to cede to rivals, coaching will start. Till then, silence.
The Jane Avenue Downside
Even with out massive tech, the buying and selling narrative faces a structural wall. Any technique constructed on a publicly obtainable mannequin is, by definition, obtainable to everybody — together with institutional quant corporations.
Qureshi’s level is easy. If a primary Claude bot can discover worthwhile trades on Polymarket, Jane Avenue can run 5,000 of them concurrently. The agency has sooner infrastructure and deeper capital. It might probably scale any worthwhile edge to zero earlier than a retail dealer even logs in. “If it’s within the uncooked mannequin, Jane Avenue is doing it proper now,” he stated.
The one manner a retail bot wins is with novel indicators absent from the bottom mannequin. A Claude occasion pointed at an API just isn’t that.
Why ‘Go Make Cash’ Doesn’t Work
Qureshi prolonged the argument past buying and selling to the broader fantasy of autonomous AI brokers incomes revenue on their very own.
The primary possibility is getting employed — having the AI agent promote its labor. However that is economically not possible. Thousands and thousands of equivalent Claude situations exist. None has a novel talent or location benefit. Hiring an AI agent is simply shopping for Anthropic compute with additional steps. No rational purchaser would pay above Anthropic’s personal API value for a similar output.
The second possibility is beginning a enterprise. This sounds extra promising, however Qureshi argued it fails for a subtler cause. Each AI agent attracts concepts from the identical pool of coaching knowledge. The result’s that all of them converge on the identical generic plans. Ask ten Claude situations for a startup thought, and also you get ten variations of the identical pitch.
Actual entrepreneurship, Qureshi stated, requires what Peter Thiel calls “earned secrets and techniques.” These are insights born from particular experiences in particular locations at particular instances. Bankless constructed its model as a result of its founders had a novel mixture of crypto experience, storytelling, and neighborhood intuition. They’d it at precisely the fitting second. A freshly spun-up Claude has no life expertise to attract from. It has no earned secrets and techniques.
This results in an uncomfortable conclusion. AI brokers can’t win at buying and selling. They can’t get employed. They can’t generate authentic enterprise concepts. So, the place is their real benefit over people? Qureshi’s reply was intentionally provocative: crime. This isn’t a future Qureshi welcomes. It’s the place the logic leads once you take away each institutional guardrail.
What This Means
The merchants constructing Polymarket bots are actual. Some earnings could also be actual, too — for now. However institutional quant corporations will arbitrage away any alpha within the base mannequin. Massive tech is not going to prepare on crypto till pressured by competitors. And the autonomous agent economic system could discover its first viable mannequin past legislation enforcement’s attain.
For the common dealer studying headlines about AI bots minting hundreds of thousands, the takeaway is implicit. The home all the time wins. In AI buying and selling, the home runs 5,000 bots with sub-millisecond latency.
The publish Why No Skilled Would Suggest AI Buying and selling Bots appeared first on BeInCrypto.
