AI agents just made their first crypto transaction on Coinbase

AI can draft your emails, generate your code, and write poetry in seconds. But ask it to book your flight or boost a social post with a $20 ad, and it hits a brick wall. Banks simply won’t issue a credit card to a cluster of Python scripts.

Now, that limitation is beginning to crumble. Coinbase CEO Brian Armstrong revealed on August 30 that the Coinbase Developer team witnessed its first crypto transaction executed entirely between two autonomous AI agents.

Tokens buying tokens

According to Armstrong, one AI agent used crypto tokens to interact with another AI agent to buy AI tokens, which are essentially data strings that let machine learning algorithms train on fresh information.

As Armstrong put it: they literally used tokens to buy tokens.

This test comes just weeks after Armstrong publicly advocated for LLMs powering chatbots like ChatGPT and Claude to get their own crypto wallets. In his view, the biggest barrier stopping AI agents from getting real, useful work done right now is their complete inability to transact.

Whether it is booking travel accommodations, purchasing raw compute, or paying for social media reach, practical everyday workflows require money. Without a checkout flow, autonomous software remains crippled.

Skipping the banking system entirely

AI agents cannot pass traditional bank KYC or open a checking account, but generating a cryptographic wallet takes a split second. Armstrong pointed out that agents can now use USDC on the Base network to settle instantly and globally with humans, merchants, or fellow software bots.

Coinbase is not alone in chasing this machine-to-machine economy. Blockchain development firm Skyfire launched a dedicated payment platform in August to allow AI agents to spend money independently.

Back in June, Web3 infrastructure firm Biconomy rolled out its Delegated Authorization Network, which co-founder Aniket Jindal explained acts as an authorization layer allowing users to safely delegate onchain trading activities to autonomous bots.

Why it matters

Connecting crypto rails to AI sounds like buzzword soup at first glance, but we think it represents a genuinely pragmatic bridge for autonomous software.

For developers and power users, giving an agent financial autonomy is the missing link between simple conversational assistants and end-to-end task execution. An agent scanning data feeds can autonomously pay micropayments for proprietary research, stitch together APIs, and deliver a finished brief without waking you up to enter a credit card number.

For the industry, it positions low-cost Layer 2 networks like Base and stablecoins like USDC as the de facto financial plumbing for the emerging agentic economy. Traditional banking rails are far too sluggish and fee-heavy to support autonomous cent-level microtransactions.

That said, the approach comes with glaring risks. AI agents hallucinate and fall prey to prompt injections constantly. The moment software controls capital, a runtime error is no longer just a weird chatbot response; it is real capital lost forever. Until sandboxing and hard budget guardrails become airtight, giving software a wallet remains a high-wire act.