AGI or not, OpenAI just gave AI agents a much better pair of hands. The more they can do, the less sense it makes to give them unlimited authority over money. Today, OpenAI released GPT-6 Astra and began a phased rollout. It can browse, use computers and carry out multistep professional work. AI is moving from explaining the next step to taking it. For payments, that changes the question. The hard part is no longer getting an agent to click “pay”. It is deciding what the agent is allowed to do with money, and making those limits hold. Among the networks supported by Remi, Sui offers mechanisms such as Payment Intents, which can limit an agent’s authority by amount, time and recipient at the blockchain layer. For a bank, those on-chain limits are only the beginning. A payment instruction still needs structured transaction information. The bank needs to see where it is in the workflow. Operations teams must reconcile the records, while compliance and audit teams need evidence they can follow later. Those limits also have to survive the move into the bank’s operating environment. That is the less visible but more practical part Remi handles: connecting programmable stablecoin payment flows to existing bank operations without asking banks to abandon familiar systems and controls. Remi coordinates transaction workflows, carries structured information, keeps status visible, and supports reconciliation and auditable records. Approval, compliance decisions and control of funds remain with the licensed institution. Making an agent pay is the demo. Making the payment fit a bank is the infrastructure. The agent can act. The institution still sets the rules. #GPT6Astra #AgenticPayments #A2A

Tokenising a deposit is easy to explain. Getting two banks’ systems to agree on who owes what is the part that makes it usable. HSBC and Standard Chartered recently completed a live cross-border transaction using tokenised deposits and Swift’s blockchain-based ledger. Each bank kept its own tokenised-deposit infrastructure. Swift’s ledger carried the payment messages, matched and netted the obligations, and final settlement still went through existing systems. So this was not one giant new system replacing everything underneath it. It was two bank systems finding a common language for one transaction. That is the useful part of the story. Bank-grade digital money will not get very far as a collection of private islands. Different systems have to work together while each bank keeps control of its own records and responsibilities.

“Sustainable trade finance” sounds straightforward until a bank has to assess an actual transaction. On 27 August, @BAFT (Bankers Association for Finance and Trade) released its Trade Finance Sustainability Guidelines for Transaction Banking. They address a practical gap: much of the existing work on sustainable finance has focused on investment banking, term lending and asset-based approaches. Trade finance is different: short-term transactions, high volumes, multiple parties and multiple jurisdictions. Then the awkward questions begin. What actually qualifies as sustainable? What evidence is enough? Who records the decision, and can someone reconstruct it later? Those are the questions we spend our time on. Infrastructure should not make the judgment for the bank. It should keep transaction information structured, status visible, reconciliation manageable and the audit trail intact while the bank applies its own due-diligence rules. BAFT’s document is guidance, not a new regulation. It still has to survive contact with a real workflow before it becomes useful.