I think “trustless” has been oversold. We solved how machines can execute agreements. We haven’t fully solved how machines decide what an agreement means when the outcome is ambiguous. Episode 2 of Agent Tank made that distinction much clearer to me. Take a simple agent-to-agent transaction. Agent A commissions Agent B to complete a task under several conditions. B delivers something. A says it doesn't satisfy the requirements. Now what? The smart contract can verify signatures, balances and predefined conditions. But if the disagreement is about whether the work actually satisfies the intent of the agreement, the problem is no longer execution. It is adjudication. And every system that handles adjudication has a trust surface. A centralized operator can be pressured. A single oracle can be wrong or manipulated. A single model can confidently misunderstand the evidence. A voting mechanism can simply turn disagreement into a contest of influence. None of these automatically produces a neutral verdict. For an agentic economy to scale, I think the missing primitive is a credible resolution process for outcomes that cannot be reduced to deterministic rules. That matters because agents will eventually negotiate contracts whose outputs are contextual: research quality, task completion, compliance with requirements, delivery conditions, external events and countless other things that are difficult to encode as simple true/false logic. This is where @GenLayer adjudication layer becomes interesting to me. The value isn't that it magically knows the truth. The value is creating infrastructure for independent reasoning to evaluate contested outcomes and reach a verdict that can be used by the protocol. That is a fundamentally different problem from building a better agent. Better agents increase what machines can do. Adjudication determines whether machines can reliably transact with each other when they disagree. And that distinction may become one of the most important pieces of infrastructure in the agentic economy. The Agent Tank hackathon is an interesting place to see what builders do with that primitive: https://t.co/zTOJ9OnqRy My takeaway from Episode 2: The moment agents start transacting autonomously, “who decides?” becomes just as important as “who executes?”