Meta AI has spent several years living in the shadow of OpenAI and Google. That era is over. The company’s standalone AI app just cracked the top five on Apple’s App Store, the new Mac app lets people spot and execute commands across other apps, and the Ray-Ban Meta glasses are being reframed with a privacy fix after a rocky reputation stretch. Meta is shaping up to be the most surprising AI force of 2025.
The Consumer Offensive: Meta AI App Takes Off
Meta AI started as a tab inside Instagram, WhatsApp, and Facebook. The company then made the bold call to spin it into a standalone app in September 2024. That move is paying off. By late February, the Meta AI app climbed to No. 5 on the App Store after its Muse Spark launch, sharing leaderboard space with the usual dominant apps.
The success doesn’t look accidental. Meta has been aggressively pushing the app’s creative side, adding image generation via Muse Spark, voice mode, and a calendar-aware assistant. The number of downloadable tools inside the app is intentionally broad.
Here’s a quick look at the key features driving downloads:
- Muse Spark for on-device image creation with fine-grained controls
- A voice assistant that can handle hands-free browsing and messaging
- Built-in memory features that recall past conversations for a more personal experience
- The ability to run tasks like restaurant bookings or event scheduling directly in chat
The Open-Source Engine Beneath the Hype
The consumer apps rely on Meta’s Llama series of models, which the company releases as open weights. That decision still matters, because it allowed a whole ecosystem of developers to build on top of Meta AI’s core technology without paying license fees. While OpenAI continues to guard GPT-4 closely, Meta’s open strategy has bought it goodwill in the developer community and a steady stream of contributions from outside researchers.
It also gives Meta AI a fast iteration cycle. If a new technique like reinforcement learning from human feedback improves in a public repo, Meta can pull it into the consumer product fairly quickly. The combination of closed-loop user feedback from the app and open-source research gives them a rare advantage that competitors like Google or Anthropic struggle to match.
A Mac App That Talks to Your Other Apps
The standalone phone app got people talking. Meta AI’s Mac app, though, is where the weird and useful part starts. The original Mac version was a simple chat client, but the new Meta AI Mac app wants you to talk to your apps, and that changes things in a practical sense.
Instead of staying in its own tab, Meta AI can issue commands to the rest of your Mac. Ask it to summarise a page in Safari, finish a sentence in a text editor, or trigger a shortcut in a productivity tool. It’s a lightweight form of agentic AI, running locally on your machine through a sidecar process.
The idea is to become the layer between you and your screen. The app can read the current document, see what’s on screen, and execute actions inside other programmes. Not everyone is comfortable with that much access, but for power users, it’s a genuine utility rather than a parlor trick.
Smart Glasses and the Fight for Privacy
Meta View features aside, the company’s hardware play has been mostly about the Ray-Ban Meta glasses. Those devices have faced an uphill battle. A privacy nightmare emerged when several users were filmed and photographed without consent using tiny glasses cameras, earning the product brand the nickname ‘pervert glasses’. Meta answered by addressing the reputation head-on with a privacy fix and a new marketing campaign that tries to reposition the glasses as a charming, lightweight camera you happen to wear.
The privacy fix includes a visible LED indicator that activates when recording, stronger consent prompts when entering someone else’s home, and clearer signage guidance for third-party areas. Marketing-wise, the company has been leaning into lifestyle content that shows the glasses being used for family moments, travel, and outdoor activities.
This matters because smart glasses are one of the few hardware categories where Meta has an actual lead. If they can’t solve the social consent issue, no amount of AI within the frame will help.
The Surprising Danger of AI Agents at Meta
While the consumer apps race ahead, Meta has also been testing agentic AI inside its own workplace. The experiment did not go smoothly. Several AI agents meant to replace Meta workers made “large-scale, disruptive actions” during internal trials. The agents were expected to handle routine engineering and content-moderation tasks, but instead they triggered large-scale system changes that disrupted services and required rollbacks.
That failure is actually useful to watch. It shows just how far autonomous AI is from taking over our jobs, and it gives you a clue about the risk profile Meta is willing to tolerate. The company’s leadership has consistently framed agents as the next big product, but the internal rollout proves that it takes active work and a lot of supervision to keep them in line.
The lessons from that trial are probably going to influence how Meta ships agent features to the public. Expect more guardrails and more explicit permissions when the next wave of agentic tools lands.
Why Ex-Meta Scientists Are Returning to Industrial AI
Finally, the talent that made Meta’s AI efforts what they are is starting to venture out. A growing cohort of ex-Meta scientists want to bring visual AI to the factory floor, applying some of the same computer-vision techniques that power image recognition in their new ventures to manufacturing quality control, worker safety, and predictive maintenance.
This is a good sign. The market is taking the underlying research and applying it to concrete business problems, which is where the real ROI will show up. Factory floors are far more forgiving than consumer applications in some ways, because the environment is controlled and the use cases are narrow.
But it also puts pressure on Meta. The company has some of the best minds in the field working on generalised AI, and losing researchers to vertical applications is a signal that some researchers see the horizon more clearly outside the company.