VQV Signal

AI AT WORK WATCH TECHNICAL

Uber compacts Claude Code sessions at 400K tokens, per a post on X

According to a post on X by @rohanpaul_ai, Uber compacts engineers' Claude Code sessions at 400K tokens, even on million-token models. The post says weekly agent requests grew 9.4x from February to August and weekly users of its AI coding tools grew 7x, while overall AI spending has stayed roughly flat since April. It also says agents reach MCP servers through command-line tools tied to a central gateway, because preloading descriptions for 100-plus tools costs 50K to 70K tokens. The source snippet is truncated, so the full details are not available here.

Source: Scobleizer News Pulse (X) · x.com Found via Robert Scoble’s Scobleizer News Pulse · a post on X, not independently verified Published 2026-11-20T02:40:00+00:00 Detected 2026-11-20T19:01:47+00:00
View original source

According to a post on X by @rohanpaul_ai, Uber compacts engineers' Claude Code sessions at 400K tokens, even on million-token models. The post says weekly agent requests grew 9.4x from February to August and weekly users of its AI coding tools grew 7x, while overall AI spending has stayed roughly flat since April. It also says agents reach MCP servers through command-line tools tied to a central gateway, because preloading descriptions for 100-plus tools costs 50K to 70K tokens. The source snippet is truncated, so the full details are not available here.

AI-assisted summary based on the listed source.

Uber compacts engineers' Claude Code sessions at 400K tokens, even on million-token models. Weekly agent requests grew 9.4x from February to August and weekly users of its AI coding tools grew 7x, while overall AI spending has stayed roughly flat since April. Agents reach MCP servers through command-line tools...

For teams running AI coding agents, token limits and tool-loading costs can shape how much work a single session can do. The post suggests usage is growing much faster than spending, which puts more pressure on cost control and on how tools are connected. Watch for whether Uber or other companies publish fuller details on these limits and gateway designs.

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

Signal Strength 73% Technical label WATCH Public Interest 44 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 73 Practical Impact Score 28 Novelty Interest Score 48 Consequence Score 22 Curiosity Score 16 Shareability Score 58

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