Why a two-year window

2030 is the horizon. But two years is close enough to be falsifiable. If we are wrong by a lot in 2028 we will say so on this blog. If we are right, we will point back to this post.

Prediction 1: One or two transactional protocols dominate

The same way TCP/IP crowded out SNA and XNS. Agent commerce cannot survive with ten incompatible transactional layers, and the natural consolidation pressure is strong. We think MCP wins on reads and that the transactional layer sits on top.

Prediction 2: 5–10% of consumer transactions are agent-initiated

This is lower than the hype and higher than today. The bottleneck is trust and dispute resolution, not model capability. The first categories to flip will be high-volume, low-ticket, low-risk: subscriptions, transport, food, routine appointments.

Prediction 3: Platform take-rate is 1–3%

Open protocols push take-rates to utility levels. The “marketplace tax” of the 2010s — 15–30% on services, 10–20% on goods — is untenable once agents can route around it. Our 2% target is deliberately in the middle.

Prediction 4: Regulation arrives on consent and disputes

The UK ICO, the EU AI Office and the US CFPB have all hinted. Expect consent-token requirements, mandatory dispute SLAs and audit-log standards by 2028.

Prediction 5: Receipt chains outlive their issuers

Content-addressed receipts will be treated as permanent evidence. Dispute tribunals will expect them. Services that cannot produce one will be uninsurable.

What we will not predict

Which LLM lab wins. Agent commerce is model-agnostic.
Which country legislates first. We will react either way.
Whether an on-chain receipt layer dominates. Unclear.

Further reading: day-in-the-life for 2030 and five myths about agent commerce.