Sluice Flow

The scariest sentence in agentic payments is "the AI decided to pay for it." Not because agents shouldn't spend — because most implementations let the model's output be the authorization. The LLM says pay, so the money moves. That's not autonomy, that's an unsecured endpoint with a personality. We built @Sluiceflow's Agent Console around a hard separation: You give the agent a budget and a plain-English policy — "only pay for AI, agent and stablecoin sources under $0.003." The model parses that into structured rules. Then deterministic code enforces those rules at the payment layer. The model recommends. Code decides what's allowed to move. So when the agent runs, every outcome is one of three things, and every one of them is on the record: Paid — relevant, within policy, worth the price. Skipped — evaluated and rejected, with the reason attached. Capped — genuinely useful, but it would breach the budget. Passing. That third one is where most systems quietly fail. An agent that can't say "this is good and I still can't afford it" isn't operating on a budget — it's operating on hope. Budget in, judgment shown. #BuildonArc @circle

The Arc Brand Guidelines and Partner Toolkit is live. For teams building on Arc or preparing partner materials, the toolkit includes guidance on logo usage, naming, example applications, and approvals. Your brand leads. Arc is the infrastructure. Explore the toolkit: arc.io/brand-guidelin…

Tried @sluiceflow on my phone today — asked their research agent a question and watched it pay each source it cited. Actual USDC, on Arc, receipts and all. Took me maybe two minutes from sign-in to seeing my first settlement. Recorded it so you can see for yourself 👇