VQV Signal

AI AT WORK WATCH TECHNICAL

Post on X says agent sandbox clusters are heavily oversubscribed because coding agents mostly wait

According to a post on X by @michielmv, self-hosted boxd clusters show heavy oversubscription, with the most packed free cluster promising 8.5 times more vCPUs to workers than its machines have (850%) while running only 13% busy. The post says the busiest paid cluster ran at 433%. The author's explanation is that coding agents spend most of their time waiting on model calls and tools, so their CPUs sit idle. The post states these are the author's own figures.

Source: Scobleizer News Pulse (X) · x.com Found via Robert Scoble’s Scobleizer News Pulse · a post on X, not independently verified Published 2026-11-19T17:35:00+00:00 Detected 2026-11-20T00:01:43+00:00
View original source

According to a post on X by @michielmv, self-hosted boxd clusters show heavy oversubscription, with the most packed free cluster promising 8.5 times more vCPUs to workers than its machines have (850%) while running only 13% busy. The post says the busiest paid cluster ran at 433%. The author's explanation is that coding agents spend most of their time waiting on model calls and tools, so their CPUs sit idle. The post states these are the author's own figures.

AI-assisted summary based on the listed source.

Michiel V shared oversubscription numbers from self-hosted boxd clusters: the most packed free cluster had 8.5 times more vCPUs promised to workers than its machines have (850%) while running only 13% busy, and the busiest paid cluster ran at 433%. His explanation is that coding agents spend most of their time...

Teams running coding agents on shared sandbox infrastructure may be able to pack more workloads onto the same machines than they assume, if the idle-time pattern holds. Anyone planning capacity or pricing for agent sandboxes should watch whether other operators publish comparable utilization data. The figures come from one person's self-hosted clusters, so they should be checked before being used as a general benchmark.

Teams using AI at work may want to compare this against current productivity and review workflows.

Signal Strength 81% Technical label WATCH Public Interest 33 Category AI AT WORK Reader Depth TECHNICAL

Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.

Public Interest components
Recognizable Entity Score 0 Practical Impact Score 46 Novelty Interest Score 70 Consequence Score 22 Curiosity Score 16 Shareability Score 49

VQV surfaced this signal because it is recent, relevant to AI Coding Tools, connected to Scobleizer News Pulse (X).