In partnership with
IN THIS ISSUE
Welcome back to The AI Field.
Anthropic is pushing long-running AI work forward while OpenClaw and OpenAI expose the control and dependency problems around it.
-Claude Fable 5.1 and Mythos 5.1 launch for advanced coding and knowledge work
-OpenClaw 2.0 adds persistent sessions and tighter controls
-OpenAI and Cursor prepare for a direct model-supply split on November 12
Read time: 5 minutes
100+ Claude Code hacks to ship code 10X faster
Top engineers at Anthropic and OpenAI say AI now writes 100% of their code.
If you're not using AI, you're spending 40 hours doing what they do in 4.
These 100+ Claude Code hacks fix that and help you ship 10x faster.
Sign up for The Code and get:
100+ Claude Code hacks used by top engineers — free
The Code newsletter — learn the latest AI tools, tips, and skills to code faster with AI in 5 minutes a day
RECENT UPDATES
MAIN STORY
Anthropic launched Claude Fable 5.1 and Claude Mythos 5.1 on September 1. Fable is generally available. Mythos offers the same capabilities with different safeguards and is available by invitation only through Project Glasswing.
Fable keeps its base API price at $10 per million input tokens and $50 per million output tokens. Cache reads fall to $0.25 per million tokens, 75% below Fable 5. Anthropic estimates that typical token-billed workloads will cost about 25% less, with savings of up to approximately 45% for highly agentic work.
The model also adds readable progress updates between tool calls, per-message effort controls, and stronger long-running coding and research. It supports a one-million-token context window and up to 128,000 output tokens.
The AI Field: The useful upgrade is not another benchmark crown. Anthropic is trying to make its strongest public model cheaper to keep running for hours. For most work, Anthropic still recommends Opus 5, so teams should move up only when their own tests justify the higher base price.
MAIN STORY
OpenClaw announced OpenClaw 2.0 late on August 30, introducing its largest update yet. Version 2026.8.1 rebuilds the agent's setup and control layer with shared sessions, private credential requests, and narrower permissions for recurring tasks.
Users can start with an existing ChatGPT or Claude subscription, an API key, or a local model. The rebuilt browser app now handles setup inside the conversation, reducing the manual configuration needed before the agent can begin work.
The key changes:
Sessions can move to paired devices or cloud workers without losing the workspace.
Progress cards show what the main agent and its subagents are doing in real time.
Masked credential requests keep passwords and API keys out of the conversation and model context.
Recurring permissions can be limited to one exact operation, reviewed later, and revoked.
The release addresses a central weakness in agent software: capability has grown faster than user control. OpenClaw now makes it easier to see what an agent is doing, where it is running, and which actions it is allowed to repeat.
Those controls do not remove the risk. OpenClaw can still access files and accounts, so users remain responsible for every permission they grant.
The AI Field: The most important upgrade is not a smarter model. It is a clearer operating layer around the model. Persistent agents will become useful at scale only when ordinary users can understand, limit, and reverse their actions.
03 · AI CODING
OpenAI says it intends to wind down the contract that supplies its models directly to Cursor after SpaceX acquired the coding company. The proposed cutoff date is November 12, and OpenAI says future models will not be provided through the agreement.
OpenAI's stated reason is trust. It says previous Musk-owned companies broke contracts and that it cannot be confident SpaceX will follow its model terms. The company is using a cancellation window triggered by Cursor's change of control.
Cursor CEO Michael Truell said the companies are still talking. Anthropic said it plans to increase compute for Claude in Cursor.
The AI Field: This is bigger than another Musk versus Altman fight. Cursor built on outside models, and one supplier is now using ownership as a reason to leave. If a critical workflow depends on one model contract, portability is not a technical detail. It is insurance.
THE AI TOOLBOX
TRENDING RIGHT NOW
Six tools worth ten minutes each. Click the name to open it.
📱 x1: Describe the iPhone app you want and it builds it screen by screen, then preps the App Store listing.
🎬 Topview Motion Studio: Turns a product brief into a finished launch video in about ten minutes, without opening After Effects.
✂️ ChatCut: A desktop video editor you drive by describing the cut you want instead of dragging clips.
📄 PageIndex: Ask one question across a stack of long documents and get an answer with a clickable citation.
🎙️ Ulpaso: Free open-source Mac app that transcribes meetings on your own machine into an editable Markdown file.
🥗 Diet Claude: Free Chrome extension showing how much of your Claude session is left and exactly when the limit resets.
QUICK HITS
A federal judge rules the Pentagon's Anthropic blacklist unlawful. The court found unlawful retaliation, due-process failures, and no evidence that the broad restrictions addressed a specific national-security risk. The government can still appeal.
Australia's ARIA Charts exclude fully AI-made songs. The rule still allows substantially human-made music that uses AI, drawing the line at core human authorship rather than any use of the technology.
Tencent open-sources its Hy4 preview model. The 770B-parameter mixture-of-experts model activates 49B parameters per token, supports more than one million tokens of context, and is live in WorkBuddy, CodeBuddy, and APIs.
MORE FROM THE FIELD
WHAT TO WATCH · NOVEMBER 12
OpenAI's proposed Cursor cutoff will test how portable modern AI products really are. Cursor and OpenAI are still talking, while Anthropic plans to increase compute for Claude inside Cursor.
The better question is whether teams can switch providers without rebuilding prompts, evaluations, and workflows. If that process is painful, the model was never the only lock-in.
YOUR TURN
Your feedback shapes what we cover next. Pick the answer that fits best.
What did you think of today’s edition?
Until next time!
Olle
Founder of The AI Field