Xi and Trump have given their officials something useful to do. The question is whether they will let them keep doing it after the next disagreement.
That is the case for giving Xi’s September 23–25 Washington visit provisional credit. Limited cooperation can make a rivalry less dangerous even when neither side trusts the other. It can also disappear the moment a leader finds confrontation more convenient. The arrangements announced after this visit deserve scrutiny on both counts. The Chinese outcomes statement
Take the tariff package. The approved lists cover roughly $30 billion of imports on each side, valued using 2024 trade. That is the value of goods covered, not $60 billion in savings. Each government still has to implement reductions through its own domestic processes. Until then, a firm has a promising announcement rather than a duty rate it can use. The framework’s terms
The Board of Trade does at least give officials a place to pursue that work. Deputies have assigned responsibilities and a schedule providing for meetings at least quarterly, though the frequency can change. A recurring negotiation has some value: the participants need not wait for their presidents to arrange another summit before addressing a practical dispute. Whether they have enough political backing to resolve it is another matter. The board’s working procedures
There is a similar bargain on communication. Washington announced an AI dialogue, under its preferred “super intelligence” terminology, and agreement to establish an incident channel. Beijing also announced an AI channel and said the militaries intend to conclude a crisis communication and prevention memorandum. The public commitments establish work to be done; they do not demonstrate that the systems already function. The U.S. fact sheet · The Chinese statement
My preference is to put real effort into those contacts. Suppose an ambiguous incident occurs during a political argument. Who receives the warning, and who can authorise a response? A channel that works then would be worth more than another cordial appearance. That is the practical value I would look for, without expecting the two governments to resolve their competition first.
The objection is that this grants too much credit to procedure. Taiwan makes that objection difficult to dismiss. Beijing’s account of the September 24 talks records Xi pressing Washington to oppose Taiwan independence. It records a Chinese demand, not a new American concession. A conversation between two presidents should not be mistaken for a settlement of the interests of everyone affected. Beijing’s readout
Still, waiting for strategic trust before doing useful work would set an unnecessarily high entry price. Implemented tariff reductions and usable emergency contacts can stand on their own merits. The harder political commitment is to keep them working when cooperation no longer makes an attractive photograph.
Put usable tariff reductions first. A supplier can act on a duty rate; it cannot plan a three-year contract around a cordial summit. The framework generally envisages adjustments no more than annually, but that is a review schedule, not protection from wider policy changes. Count the shipments that benefit now. Reserve “stability” for an arrangement that survives a serious disagreement.
Suppose an engineer spots an AI-related disruption but cannot yet explain its cause. Do they alert the other government, or wait until they can defend every detail? That decision could matter more than whether a test message reaches the right inbox.
The announced incident channel does not yet spell out the reporting thresholds. It needs room for a warning that says what is known, what is uncertain and what is being checked. A shared reporting template could help with that. If reporting requires a polished explanation first, the other side may spend the delay inventing its own.
Someone answering the phone is the start. Can they reach an official authorised to change an operation, and can that instruction reach the units involved?
The military memorandum is still to be concluded. Test that chain with a delayed response or conflicting reports. Keep it separate from the AI channel. Even a successful exercise would show capacity; leaders would still have to choose restraint.