This is a little shocking coming from an organization as large and sophisticated as Meta, but it also makes perfect sense. The agentic AI space is moving so fast that even Zuck cannot keep up.
That should tell you something. There is a very narrow gap between an AI agent feeling like a game changer and feeling like a useless toy. Use the wrong model, run the wrong harness, skip the skill layer, give it weak tools, ignore memory, or fail to tune the workflow, and the whole thing breaks down fast. The user does not care that the underlying technology is impressive. They care whether it works.
I think this is where the real gold rush is happening right now. The winners will be the people who understand that agentic AI is less about a flashy chatbot and more about systems design. Models matter, but the harness matters just as much. Tools matter. Memory matters. Feedback loops matter.
Meta will figure plenty of this out because they have the money, talent, and distribution. But the fact that even they are openly admitting the experience has not come together as expected should wake people up. There is a massive opportunity for smaller teams who actually have this stuff working today.