Feature update Provisional 90% confidence first seen
The ttok CLI token-counting tool was updated to fix a tokenizer default that had been using the GPT-4 tokenizer instead of the intended GPT-5/GPT-6 tokenizer. The release also modernizes the project’s CI/workflow and adds a --list-models command (using OpenAI’s tiktoken) to display available model names.
The daily briefing
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.
TLDRocket
9 October 2026
The biggest thread today isn’t model releases—it’s documentation becoming operational infrastructure as AI agents start treating web content as gospel. MIT researchers and an author point out that agents make far more retrieval requests than humans browse: in August, they logged 257 million documentation requests versus 131 million human page loads. The result is predictable but nasty—agents can repeat outdated or contradictory passages because they assume retrieved text is authoritative. That frames documentation work the way IT has always treated databases: sources of truth, monitoring, structured retrieval, and explicit decisions about what the system should believe when answers conflict.
Read the full briefing →