got pulled into a meeting today. apparently we're adding an Agentic AI to the team. it will learn our environment, handle tasks autonomously, and integrate via API. it does not need onboarding, a desk, or health insurance. Great.
i have one question nobody in that meeting could answer. how does it actually work?
not philosophically. like what is the system. because from what i can tell it's an LLM with tools strapped to it, some kind of memory layer nobody can fully explain, and a control loop that lets it run without a human saying yes to every step. which means somewhere in my company's stack there is now a process with access to our tools, our data, and apparently a better performance review than me, and i genuinely do not understand the architecture.
the memory part especially. is it reading our docs at runtime, is it storing embeddings somewhere, is it getting fine tuned on our internal data. these feel like important questions. my manager said "it learns over time" and moved on to the next slide.
can someone who actually understands how these systems are built explain it to me like i'm a senior engineer who is totally fine and not at all spiraling.
[link] [comments]
