Anthropic Launches Fast Mode for Claude Code
Today's top 20 insights for PM Builders, ranked by relevance from X, Blogs, YouTube, and LinkedIn.
Anthropic Launches Fast Mode for Claude Code
#1 đť•Ź
Claude launched Fast mode for Claude Code users with extra usage enabled (via /fast) and rolled it out in research preview on @cursor_ai, @emergentlabs, @FactoryAI, @figma, @github Copilot, @Lovable, @v0, and @windsurf.
#2 đź“ť Anthropic Engineering
Quantifying infrastructure noise in agentic coding evals - Infrastructure configuration can significantly impact agentic coding benchmarks, sometimes more than the gap between top models.
#3 đť•Ź
Sundar Pichai uses Waymo’s Genie 3 to create high-fidelity, interactive simulations of rare driving events that are nearly impossible to capture in the real world.
#4 đť•Ź
Boris Cherny launched the /fast mode in Opus, using significantly more compute than Opus 4.6 and incurring higher costs for incident response and accelerated work on critical projects, and announced his team built and tested this experimental fast mode for Opus 4.6 with Claude over the past few weeks (tweet).
#5 đź“ť Simon Willison
How StrongDM’s AI team build serious software without even looking at the code - A look into how StrongDM's AI team operates without human oversight in coding.
#6 ▶️
Reverse engineer Claude Code Agent Teams
AI Jason
Demonstrates how to install and use the Cloud Code agent teams feature (v2.1.34) by enabling the experimental flag in settings.json and launching collaborative AI agent sessions with “cloud-teammate --mode.”
- Requires Cloud Code version 2.1.34 and adding "cloud_code_experimental_agent_teams": 1 to your global settings.json file.
- Use T-Max (or iTerm2 on Mac with Python API enabled) and run "cloud-teammate --mode" to open split-view sessions for each agent teammate.
- Introduces a "team_create" tool that generates a config file in doc/teams (with an empty team member array) and a "task_create" tool that writes JSON task files in doc/teams/tasks including subject, description, status, blocked, and blocked_by fields.
#7 đť•Ź
Teresa Torres launched Earmark’s “personas” feature, creating AI agents that simulate security, legal, and accessibility experts to ask role-specific questions in product meetings.
#8 đť•Ź
Tal Raviv gave Opus 4.5 read-only access to his Mercury bank account using Mercury’s MCP connector (official Anthropic app, quick OAuth) to diagnose a tax shortfall.
#9 đź“ť PromptLayer Blog
How do teams identify failure cases in production LLM systems? - Production LLM systems fail in ways that traditional software never did, and teams struggle to catch issues that are non-deterministic and context-dependent.
#10 đź“ť PromptLayer Blog
How to install OpenClaw: Step-by-step guide (formerly ClawDBot/MoltBot) - This guide walks you through getting OpenClaw running on your machine, detailing what it does and how to set it up.
#11 đź“ť Doug Turnbull
Elasticsearch hybrid search in practice - Elasticsearch knn query is both a joy and a headache - here is where you'll get stuck and the hacks I've used to overcome them.
#12 đź“ť Simon Willison
Vouch - Mitchell Hashimoto introduces a new system to manage AI-generated PRs in open source projects.
#13 đť•Ź
Peter Yang published a 30-minute YouTube tutorial on safely setting up and running the OpenClaw personal agent.
#14 đť•Ź
Santiago writes significantly fewer lines of code with agentic coding in his development workflow.
#15 ▶️
The Two Models that will Dominate AI Discussions Just Got Released (Claude Opus 4.6 + GPT 5.3 Codex)
AI Explained
Benchmark comparison shows Claude Opus 4.6 outperforms GPT 5.2 by about 140 ELO points on the GDP val white-collar work benchmark, while GPT 5.3 Codex achieves 77.3% on TerminalBench 2.0 extra-high settings versus 65.4% for Opus 4.6 Max.
- Claude Opus 4.6 outperforms GPT 5.2 on GDP val by ~140 ELO points, implying ~70% preference for Opus 4.6 outputs
- GPT 5.3 Codex (extra high settings) scores 77.3% on TerminalBench 2.0, compared to 65.4% for Claude Opus 4.6 Max
- Claude Opus 4.6 supports a 1,000,000-token context window, matching Gemini 3 Pro
#16 đť•Ź
Peter Yang identified a huge market for video editing, noting many creators pay over $1,000 per video edit.
#17 đť•Ź
Guillermo Rauch received cold emails in the past few weeks from a 15-year-old and a 16-year-old offering significant technical insights and contributions.
#18 in
Dharmesh Shah stated that reporting answers directly where the data lives boosts usage and that deterministic AI systems deliver predictable, precise answers without guessing.
#19 đť•Ź
Kevin Yien integrated an AI assistant into search (invoked via Cmd K) for AI docs search and is working on further enhancements.