AI agents may move more of the industry’s value towards the software that gives models memory, tools and room to act, analysts told the South China Morning Post. Meta’s Muse has helped draw attention to that shift, but downloads are an early sign of interest, not proof of where the money will ultimately land.
Watch Desk analysis
What happened
The South China Morning Post reports that analysts see agent harnesses as a potential route for value to move away from foundational models and towards the software layer. Unlike a chatbot that simply responds to prompts, an agent harness can provide persistent memory and tool access, allowing it to take on more complex tasks.
Meta says Muse can work across the web and connected services such as email and calendars, and help users fill out forms, make purchases and book travel while remembering their preferences. The report says Muse passed 2.5 million downloads in its first two weeks after launching on 8 September and topped the US free-app rankings on Apple’s and Google’s stores. Those figures show attention, not sustained use or revenue.
Why it matters
If people use agents to complete tasks rather than just ask questions, the software that connects models to tools and services could become an important part of the product. That raises a live industry question: will the companies supplying the underlying models capture most of the value, or will the agent layer become a business in its own right?
The distinction matters for users, too. A capable model is only part of an agent experience; memory, access to services and the ability to carry out actions shape what it can actually do. The report describes an emerging market and analysts’ view of its potential, not a settled shift in industry economics.
Our read
Agent harnesses are where the chatbot starts doing more than chatting, which is a more interesting contest than another round of model bragging rights. But downloads are a starting gun, not a finish line. The real test is whether people return, trust the tools with useful tasks and find enough value for someone to pay for the software around the model.
What to watch
- Whether Muse turns early downloads into sustained use.
- Which tasks agent products can reliably complete across connected services.
- Whether agent software companies establish a business beyond access to foundation models.
Discussion spark: If AI agents become useful, will the lasting value sit with the companies building the models or with the software that gives those models tools and memory?
Sources and evidence
- Can China match up with Meta’s Muse in race to harness AI agents? – South China Morning Post (8 October 2026, 14:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.