1 hour ago
25
Meta just gave its AI assistant a significant brain upgrade, and the architecture powering it marks a notable strategic pivot. The company announced on July 24 that Meta AI can now autonomously execute tasks, generate daily briefings, and handle recurring workflows like weekly meal plans and reminders, all powered by a new proprietary model called Muse Spark 1.1.
Here’s the thing that matters beyond the feature list: Meta built this on Muse, not Llama. That’s a meaningful departure from the open-source model strategy that made Meta a darling of the decentralized AI crowd.
From Llama to Muse: what changed and why it matters
Muse Spark 1.1 launched on July 9, roughly two weeks before Meta rolled out the task automation features built on top of it. The model was developed by Meta Superintelligence Labs and is designed specifically for what the industry calls “agentic workflows,” which is a fancy way of saying the AI can actually do things for you rather than just answering questions.
In English: instead of asking Meta AI to find a recipe, you tell it once that you want a weekly meal plan, and it handles it going forward. The AI understands your context better, requires less hand-holding, and can chain together multiple actions autonomously.
The features are currently live in select markets only, which suggests Meta is taking a measured approach to deployment.
Meta’s Llama models became the backbone of a significant portion of the open-source AI ecosystem. The pivot to proprietary Muse models for Meta’s core consumer product signals that the company sees its competitive future in closed, optimized systems rather than open-source community goodwill.
The crypto AI angle: open-source headwinds
The entire AI token sector, which includes projects focused on decentralized compute, AI agent frameworks, and inference marketplaces, has built much of its narrative on the assumption that open-source AI models would continue to improve and proliferate. Llama was Exhibit A in that thesis. When the world’s fifth-largest company by market cap releases powerful models that anyone can run, it creates natural demand for decentralized infrastructure to host them.
Meta shifting its flagship product away from Llama doesn’t kill the open-source AI narrative outright. Llama models still exist and are widely deployed. But it does introduce a question: if the company that championed open-source AI is now building its most capable systems behind closed doors, what does that mean for projects whose entire value proposition depends on open model availability?
What this means for investors
From a traditional markets perspective, Meta’s push into autonomous AI functionality is straightforward bullish positioning. Task automation that reduces friction and increases daily engagement with Meta AI directly supports the company’s advertising revenue model.
The projects most insulated from this dynamic are those focused on infrastructure layers, like decentralized GPU compute and verifiable inference, rather than consumer-facing AI agents that compete directly with Meta’s offering.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·