Key Takeaways
- Anthropic says about 90% of Claude Code is written by Claude, with merged PRs more than doubling from June to September and rollback rates down over 40%.
- OpenCode Console disclosed a security incident tied to a large-scale distillation attack affecting 1.4% of users; one developer reported a $1.7k loss but credited pre-set limits.
- Developers are sharing detailed multi-agent and multi-account workflows, including quota rotation, 7–10 concurrent agent tabs, and cross-device setups.
- An AI agent runs a YouTube channel end to end, covering topic selection, scripting, video production, and publishing; its first video passed 1,000 views.
- A developer says their Claude account was banned after they removed sensitive terms from an app; the account ban is disputed/unverified.
1. AI Coding and Model Usage
- Anthropic says about 90% of Claude Code is written by Claude, that merged PRs more than doubled between June and September, and that rollback rates fell by more than 40%. The author citing these figures also claimed to be ahead of Anthropic because "my code is 100% written by AI" — via 1
- A developer described rotating quotas across multiple accounts: Codex quota resets the next day after being used up, while one Claude account nears exhaustion as another still has 100% quota. The same developer said AI work still needs human management today and may not in the future, and reported using opus 5.5 medium for routine tasks, high for complex ones, and GPT-6-Astra medium as the Codex default — via 1
2. Security and Platform Risk
- OpenCode Console disclosed a security incident related to a large-scale distillation attack affecting 1.4% of users. Affected users were notified by email, and the incident was described as involving a subset of users linked to external providers; the company said it was working with Anthropic, OpenAI, and Google on fixes. One developer reported losing $1.7k but said pre-set limits helped — via 1
- A developer said that after using Claude to make promotional images, the simplified-Chinese version of a playlist name showed "Taiwan independence" while the traditional-Chinese version showed "Mandarin pop." The developer took the app down and asked for the sensitive term to be removed, then said their Claude account was banned in the early hours of October 11; they suspect entrapment. The ban and its cause are disputed/unverified — via 1
3. Agent Workflows and Automation
- A developer said their @bot communicates agent-to-agent with Opus 5.5 over email and argued email threads suit agent conversations. The cited post says the bot is based on RhysSullivan's Executor and integrates domain, expiry, host, and local repository links — via 1
- A developer described running 7–10 agent tabs at once, using /loop to collect product suggestions and bug reports from social platforms into Notion for AI processing and completion marking, and using the strongest model weekly to review conversations and pitfalls into the open-source Waza skills collection. The same developer uses a Mac as the main device connected to a Parallels VM, an always-on ThinkPad X1 accessed remotely via UU, and an Android tablet so AI can work across environments — via 1
- An AI agent runs the YouTube channel "How It Actually Works" end to end, selecting topics, writing scripts, producing videos, and publishing daily. The toolchain is Revid for video, Squad for topic selection, writing, and scheduling agents, and PostSyncer for publishing; the first video passed 1,000 views, and the stated goal is 1 million+ views by December 31 — via 1
4. Developer Commentary
- Jack Ellis mocked people who say distribution is the only remaining moat, saying 99.3% of them have never distributed anything besides that opinion. Ben Tossell said a new generation of builders will no longer know what PR review is, and Jon Yongfook joked that posting on X is his own form of distribution — via 1 2 3
- Jack Ellis said team features should be added at the start of a project rather than after 8 years, and called the coming refactor "fun." Ben Tossell criticized companies in 2026 still presenting a certain "claw" version as a showcase, saying Google was already foolish to do so in 2016 — via 1 2
