Discussion about this post

User's avatar
Arturo Barrantes's avatar

I’m in between two options:

Excited about the possibilities

Fascinated by the weirdness of it

I believe it could work, but we’re still far from that reality. Here’s why:

Right now, we have to implement tools for everything agents are supposed to do. For full autonomy, either we’d need to code every possible capability, or allow the agent to build its own tools. That would be interesting—but without proper guardrails, who knows what could happen?

Agentic AI still faces significant drawbacks: persisting memory, processing huge numbers of tokens, and the degradation of decision quality over time because agents lose track in longer conversations.

Despite these challenges, progress over the last two years has been tremendous. So, what you describe in the post might become real—but I think it’s still at least 5 years away.

Expand full comment

No posts

Ready for more?