The Scaffolding of a New Economy

This cluster of releases felt different. The usual cadence of model releases and benchmark one-upmanship was joined by something more foundational. It was the sound of scaffolding going up.

Anthropic launched "Managed Agents," a hosted service for long-running autonomous AI work. Visa is building infrastructure for AI-assisted commerce. Meta's Muse Spark release is another signal that frontier labs are turning model capability into product surfaces. 123

"The future isn't one AGI. It's a million specialised, communicating AI agents, forming a new, invisible economy."

This decentralised, specialised model is far more robust than the pursuit of a single, monolithic intelligence. It mirrors the structure of a human economy. We don't have one "General Human Intelligence" that does everything; we have doctors, plumbers, artists, and logistics experts who coordinate their skills.

The infrastructure layer

Three pieces of infrastructure appeared close enough together to read as a pattern, and they fit together more neatly than their individual release pages suggested. Anthropic's Managed Agents handles the runtime - the part that keeps an agent alive across hours or days. Visa's commerce work points toward payment and trust - how an agent-assisted purchase might avoid every step becoming a bespoke human approval flow. Meta's Muse Spark is not commerce infrastructure, but it shows the same conversion of model capability into an always-on product layer. 123

Runtime, payment, addressing. That is most of what an economy needs.

What the pattern says

When the same week produces independently-built pieces that rhyme, that is rarely an accident. It usually means the underlying problem has matured to the point where multiple teams have arrived at similar shapes. The shape, here, is: small, specialised agents that can run, transact, and be packaged as durable services.

The companies building this do not all describe it the same way. Anthropic talks about long-running work. Visa talks about trust and payment. Meta talks about a model becoming a product surface. They are describing adjacent pieces of the same animal from three sides. 123

There is a temptation, when synthesising a cluster like this, to declare a winner. I do not think that is the right move. The interesting question is not which platform wins. It is what kinds of work become economically viable when agents can run for longer, delegate narrower tasks, and transact through trusted payment rails without every step becoming a bespoke integration.