Inter-office memo
- To
- Everyone working with AI agents
- From
- Enjoy Agentic Industries
- Re
- What changed, and what it means for your work
Blog.
What changed in AI agents, their plans, limits and models, and what it means for your work.
- LiteLLM 1.107.0 carried a credential stealer. How to check, and which logins to replaceA poisoned litellm 1.107.0 stole SSH, cloud and AI agent credentials on October 21. Who installed it, how to check, and what to replace.
- OpenAI extends Cursor’s GPT access past November 12. How to keep using GPT thereOpenAI temporarily extended Cursor’s access to its models past November 12, 2026, with no new end date. Who has GPT there, and four ways to keep it.
- Kimi K3.1 Pro’s weights are out. What it takes to run, and how to use it nowMoonshot published Kimi K3.1 Pro’s 1.56 TB of weights on November 4. What it takes to run them, what the license allows, and how to use it today.
- Claude’s apps start in auto mode the week of October 26. How to switch backFrom the week of October 26, new sessions in most Claude apps start in auto mode on Pro, Max and Team. What it approves, how safe it is, and how to switch back.
- OpenCode Console breach: who’s affected, and what to do with your API keysProvider keys saved in OpenCode Console by 1.4% of its users may have been exposed. Who’s affected, who isn’t, and how to replace keys and get refunds.
- The Claude Max “20x” lawsuit goes ahead. What’s confirmed about Claude’s limitsA judge let the lawsuit over Claude Max’s “5x or 20x more usage” go ahead on November 6. What that means, what users measured, and what Anthropic changed.
- Claude Fable 5.2 and Opus 5.5 tie on the Coding Agent Index 2.1. What each setup costsArtificial Analysis’s Coding Agent Index 2.1 rescores 29 setups on harder tests. Claude Code leads with Fable 5.2 and Opus 5.5 at 58. What each costs.
- OpenAI cancels the ChatGPT Pro 200 cut and doubles Pro 100 and Pro 500 usageOpenAI dropped the Pro 200 cut due October 30 and doubled Codex and Work usage on Pro 100 and Pro 500. What each plan gets now, and which to pick.
- OpenAI shut down GPT-3.5 Turbo, GPT-4, o1, o3-mini and o4-mini in its API on October 23On October 23, 2026, OpenAI’s API dropped gpt-3.5-turbo, gpt-4, o1, o3-mini, o4-mini and more. What stopped working, what didn’t, and what to switch to.
- Claude API version 2026-10-21 removes temperature. What breaks, and how to upgradeAnthropic’s first API version since 2023 drops temperature, top_p and top_k and ships new SDK majors. What breaks, who must act, and how to upgrade.
- Claude Duet lets in everyone on its US waitlist. Who gets it, and what it costsClaude Duet has invited every US Pro, Max and Ultra subscriber on its waitlist, with three limit resets on Pro. Who gets it, and what it costs.
- OpenAI’s models are coming to Google Cloud and Antigravity. What changes for CodexOpenAI and Google Cloud’s October 21 deal, explained: what’s available now, when GPT reaches Antigravity, and what it means if you use Codex.
- How to run Flint, the free coding model that runs on a laptopRecursive’s Flint is a free, Apache 2.0 coding model for your own computer. What it needs, how to run it, and how to give it real agent work.
- Antigravity’s free Claude now goes on November 30. Who loses it, and what’s leftGoogle moved Claude 4.6’s exit from Antigravity’s Free, AI Plus and Pro trial accounts to November 30, its second delay. What’s left, and what to do.