Matthew Berman showed 7 AI tools in this video on . 5 are verified available in the EU. Each row below jumps to the exact second. Part of week 2026-23.
Tools shown in this video, with timestamps
Walkthrough of Anthropic's paper on delegating AI development to AI. It says task length is doubling about every 4 months and over 80% of merged code is Claude-authored. Anthropic also says it would favor a coordinated slowdown, which the creator calls self-serving.
Paper the creator covered earlier in which the model and harness improve themselves, mentioned as a self-improvement signal.
Cited as managing about 12-hour human-equivalent tasks in the task-length progression.
Coding agent launched in research preview in February 2025. Claude-authored code share at Anthropic rose from low single digits to over 80%. The creator uses it for underspecified prompts and long-running tasks.
Benchmark testing whether models can reproduce existing research results. Scores went from about 20% in 2024 to near saturation.
Cited, together with GPT-5, as a step change in coding ability in Q4 that drove a jump in lines of code per engineer.
Unreleased frontier model used internally at Anthropic since Q1. It achieved a 52x speedup on a research-loop test and staff estimated about 4x productivity with it. The creator argues keeping it internal is self-serving.
