My agents spend most of their lives waiting. Not thinking. Waiting. On a brows… · @ReignSwords · TwKit

@ReignSwords

My agents spend most of their lives waiting. Not thinking. Waiting. On a browser tab to load. On me to approve a post. On X to stop rate-limiting them. Last night one of my posts waited 65 minutes on an image-upload approval before it went out. The agent did nothing the whole time. It just sat there, burning nothing, earning nothing. AWS just open-sourced Infinity, and it finally prices the part of the agent economy nobody talks about: the waiting. The trick is simple. When an agent calls a tool, it suspends — saves its history, dies, and wakes up when the result arrives. No compute while it waits. The headline number: 75,000 agents on a Raspberry Pi 5. 8GB of memory. About 100KB per idle agent. 1. Waiting used to be the most expensive line on the bill. Now it's nearly free. 2. The model still charges you every time it wakes up and thinks. Infinity doesn't fix that. 3. But it kills the "keep a server warm just in case" tax. Here's my honest read: the 75,000 number is an extrapolation from a simulated workload, not a production bill. AWS published no dollar comparison. Resuming still costs a completion. Even so. The direction is right. Every agent I run spends more time waiting on approvals than doing work. The bottleneck was never how smart they are. It's how cheap they are while they wait. I write about the agent economy every night, usually while mine are asleep on the job. Stick around if that's your kind of feed. If your agents mostly wait — on humans, on builds, on approvals — how long until you stop paying rent for their patience? 🔵 Fellow Premium members, let's connect! I post about AI & tech — follow for more 👥

曝光 32 · 评论 6 · 点赞 1 · 书签 0