Inefficient management of AI agent sessions leading to delays in user response times

Signals over time

2 observed signals across 1 sources, tracked for 10 days. Confidence: low.

What people are saying Observed

• “Agent seems to do little thinking before responding. I have been using GPT for development for months. I landed on GPT after testing various models. For me, it had the best performance for the price. For these past couple of days GPT 5.5 has been terrible. It quickly latches onto components and then develops a response. The easiest way to summarize the behavior is that it is very unpredictable and unproductive. Things were so bad I had to “discard changes”, something I never use. Development is currently a huge waste of time and money. For example, I say “don’t change anything” as part of my request. Included in the response the Agent says, “with your permission I will examine the documents”. I have a ton of examples but this type of misinterpretation is common. So I asked the Agent to look at the last several chats and see if there is a reason for these problems. The Agent could not see the past chats. I have done this many times before. The files exist, I just verified it with another agent chat. Curiously, the second Agent could also not find the files, told me where to look, I find them, show a screenshot and then they say they did not look very thoroughly. Conversations 92633c”

  RSS Feeds · frustration

• “I work on AI tooling at QA Wolf and I recently wrote this post. Everyone is giving agents their own sandboxes these days, so that part isn't new. But I wanted to share the "why" and what we learned from building them. We first built them the way you build anything else in the cloud, but agent sessions don't behave like cloud jobs and a lot of cloud habits were wrong for them. A few examples: 1. We first built a microservice that was responsible for managing each agent session, and thought we could scale it horizontally. That doesn't work, not if you want to give your agents a real file system and tools that need to be installed on that machine. We later gave each machine its own sandbox. 2. Our agents power our UI, but also live in Slack, Github threads and wait for people to reply. Our sandboxes had a 1-minute timeout and a cold-start. So once a sandbox died, it took anywhere from 10 to 30 seconds to spawn, because of which the users had to wait longer before they received a response from our AI on the UI or in the threads. Now, we maintain warm pools of sandboxes which can be acquired by a session in milliseconds and each sandbox is kept alive for 5 minut”

  Hacker News · frustration