Theo: Agents SDK burns your 5-hour Claude limit faster
A user emptied a full 5-hour Claude Max window in about 30 minutes on one bug-fixing session. Theo spent a day on it and says it is a real bug on Anthropic's side, not your prompting.
Read the whole week in one goTheo: Gemini hallucinates in over 80% of runs
Receipts
Theo says Anthropic's Agents SDK drains 5-hour limits faster than Claude Code
Theo says he confirmed a real bug on Anthropic's side: all Agents SDK usage drains your 5-hour limits faster than using Claude Code directly, though weekly limits are unaffected. He suspects the recent boost to 5h limits was never applied to the SDK, and says he has reported it.
it emptied the whole 5-hour limit in ~30 mins
Anthropic now tells companies to delete unnecessary subagents
Anthropic is advising companies to remove unnecessary subagents, according to @cherry_mx_reds, with Anthropic's team finding that a single agent equipped with skills can frequently outperform complex multi-agent systems, and one customer seeing roughly 2x faster execution. The suggested split: skills teach how, tools and MCP do things, subagents only for parallel work, isolated context, or tasks one agent struggles with.
improving agent architecture is becoming less about what you can add and more about what you can delete
How they actually work
Theo cloned his prod data and hill-climbed search for 18 hours straight
Theo rebuilt T3 Code's search with a loop that ran about 18 hours: clone real data from the production DBs, collect example queries that returned bad results, synthesize 50 more from his own usage, then hill climb with Opus, Sol and Wonder Full until 53 of 54 cases passed.