Jack Dorsey just open-sourced a Trojan horse. Block's Berd isn't a developer tool — it's a payment terminal for the machine economy.
I've been tracking on-chain liquidity for years. I've seen hacks, flash loans, and fake floors. But this is different. This is a deliberate, strategic move by a company that processes billions in payments to plant a flag in the AI agent stack. And the market is sleeping on it.
Context: Why Now?
Block (NYSE: SQ) — the fintech giant behind Square, Cash App, and Jack Dorsey's Bitcoin obsession — dropped Berd into the open-source world with little fanfare. A desktop app for managing AI agents. Not a web platform. Not a SDK. A local-first, native application.
At first glance, it's a crowded space. LangSmith, Dify, CrewAI, n8n, Langfuse — the agent management aisle is overflowing. But none of them have what Block has: a direct pipeline to the payment rails used by millions of merchants and consumers.
This is not a coincidence. The AI agent boom is creating a new category of digital transactions — machine-to-machine commerce. Agents book flights, order supplies, subscribe to services. Who captures the payment? Right now, no one. Stripe launched an Agent Toolkit in June 2025. Block is now countering with Berd.
The desktop form factor is a signal. Web apps are for collaboration. Desktop apps are for local control, privacy, and — crucially — sandboxing. Block wants agents to run locally, to access local files, to call local tools. And then, when those agents need to pay for something, they'll leverage Block's infrastructure.
Core: The Technical Strategy Behind the Open Source
"Controlled open-source" is the label the original article used. I'll go further: it's a classic ecosystem play. Block is releasing the code but not the keys to the castle. The license likely restricts commercial use or requires a separate agreement for cloud deployment. This is not unique — Meta's Llama, Mistral, and Microsoft's Phi all do this. The goal is to get developers to build on Berd, embed it into their workflows, and then eventually need to hook into Block's payment APIs.
I've seen this pattern before. In 2017, during the EOS hypercontract race, I spent 72 hours stress-testing the beta client. The core team released code but kept the final consensus mechanism under wraps. Developers flocked to it, built tools, and when the mainnet launched, they were locked in. Block is doing the same thing — but with payment rails instead of consensus.
Berd's technical architecture is likely based on Electron or Tauri, given the cross-platform desktop requirement. The real value, however, is in the agent runtime. Local execution means agents can interact with the user's machine — files, browsers, even hardware wallets. This is a sandboxed environment, but it's also a gateway. If Berd gains traction, Block can introduce a "pay" tool call that triggers a Square invoice or a Cash App payment.
Let me break down the hidden features that the announcement mentioned but didn't emphasize:
- Multi-model support: Berd almost certainly supports OpenAI, Anthropic, and local models via Ollama/LM Studio. That's table stakes. But the differentiator is that Block can offer a "first-party" model that is optimized for transaction processing.
- Tool call logging: Every agent action is logged locally. This is a goldmine for audit trails, but it also creates a data moat. Block can use aggregated (anonymized) data to understand agent behavior and improve payment routing.
- Payment API integration: The APIs are likely already there, just not announced. Developers can build agent tools that call Square's payment endpoints. This is the killer feature that no other agent management platform has.
Contrarian: The "Controlled" Open Source is Actually a Feature, Not a Bug
The original article framed "controlled open-source" as a limitation — a sign that Block is not truly committed to community collaboration. I disagree.
In 2020, when I uncovered the Uniswap V2 liquidity hack by monitoring oracle price deviations, I learned something important: transparency is not the same as permissionlessness. The hack was possible because the code was open, but the fix required coordination. Controlled open-source allows Block to maintain a single point of accountability for security and compliance. Agent payments are not a toy. If an agent mishandles a $10,000 transaction, the user needs a company to blame, not a GitHub repo.
Block's financial compliance infrastructure — anti-money laundering, fraud detection, KYC — is a massive advantage. Stripe has it too. But LangSmith and Dify do not. Berd's controlled open-source is a signal that Block is serious about security. They want developers to innovate on the agent layer, but they want to own the critical payment pipeline.
The real contrarian angle is this: the market is underestimating the speed of agent commerce. I've been tracking institutional inflows since the 2024 Bitcoin ETF approvals. I saw how BlackRock's accumulation drained exchange liquidity. That same dynamic is about to happen in agent payments. The first company to provide a seamless agent-to-payment integration will capture a new asset class: machine transaction volume.
Block is not trying to compete with LangChain on features. They are trying to make Berd the default operating system for agent commerce. The desktop app is the bait. The payment API is the hook.
Takeaway: The Next Battlefield is Agent Commerce
The question is not whether Berd will succeed as a developer tool. It's whether Block can execute before Stripe or OpenAI build their own payment rails. My experience with the 2021 Bored Ape Yacht Club floor crash taught me that narratives can shift overnight. The same is true here.
Berd is a bet on a future where agents transact autonomously. If that future arrives in 12 months, Block wins. If it takes 5 years, the current agent management platforms may have already established their own payment integrations.
Gas up or get left behind. The machine economy is coming, and Berd is Block's ticket to the front of the line.
Liquidity is blood. Watch it drain from the pure-play AI tools into the payment-enabled platforms. Enter fast. Exit faster — but only if you're positioned on the right side of the agent commerce stack.
I'll be watching the GitHub repo for the first signs of payment API endpoints. That's the signal. That's when the real game begins.
