← sources

For those building AI agents, what's been the biggest pain in the ass so far?

source: reddit (reddit:AI_Agents)published 2026-11-10 20:42ingested 2026-11-11 02:36

retained text

Hey everyone! This is my first-ever Reddit post, so bear with me if I'm not too familiar with how things work around here.

I'm working with a few friends on an AI infrastructure project. Before we get too deep into building anything, I figured it'd be worth asking people who've actually worked with AI agents what their experiences have been like.

We've been looking into a few areas in particular:

* **Reliability:** Agents crashing halfway through a workflow, losing progress, or requiring restarts.
* **Security:** Managing what tools, APIs, or resources agents should actually be allowed to access.
* **Costs:** Agents getting stuck in loops, burning through tokens, or running up unexpected API bills.

I'm curious whether any of you have dealt with these problems, and if so, how you handled them. Did you end up building your own solutions, or have frameworks like LangGraph, Temporal, etc. already handled most of it for you?

I'm trying to obtain a better understanding of what's actually frustrating developers rather than just assuming certain problems need solving.

Appreciate any insights!

document fetched 2026-11-11 00:18 · canonical https://reddit.com/r/AI_Agents/comments/1xsy2pf/for_those_building_ai_agents_whats_been_the