Skip to content
Dustin's AI Lab
Go back

The Perpetual Token-Burning Machine: Anthropic Really Is Trying

A few scattered gripes about Anthropic's relentless push to get users burning tokens: 103 auto-triggered agents, five-layer Russian-doll subagents, and Formosa-Plastics-tier flexing.


I didn’t even turn on ultracode, and it still auto-triggered a dynamic workflow on me. By the time I looked up, 103 agents were already on their way. Infuriating. Looks like I need to figure out which parameter forcibly shuts this off. Though honestly, my request was just killing a chicken, so I can’t really tell where the ox-cleaver’s superiority lies.

Next up: they’ll probably roll out /effort xxhigh, recommended for optimal performance.

Claude Code subagents have officially entered the Russian-doll era: a subagent can call another subagent, up to five layers deep. Told you. The road to burning tokens never ends.

The hottest flex of 2026: skip the Claude subscription, hit the API directly for Fable 5 max, enable dynamic workflow plus five-layer-deep subagents, and you’ll soon command a Formosa-Plastics-tier subagent army.

When it comes to encouraging users to burn tokens, Anthropic really is trying harder than anyone before or after.


Share this post on:

Previous Post
My First All-English Two-Hour Corporate Training: I Burned the Script Two Hours Before
Next Post
The Five-Step SOP for Sneaking Claude Code Runs From the Hotel Over Dinner