AI Coding Assistants

Today's Stories

1. 1

   Anthropic launches Claude Fable 5.1, cuts costs up to 45%

   Anthropic launched Claude Fable 5.1 and Mythos 5.1, its most capable AI models yet, with gains in agentic coding and text quality. The company also cut costs by up to 45 percent for heavily agentic tasks. Alongside the models, Anthropic introduced built-in watermarks and a detection API in private preview for regulators, media outlets, fact-checkers, and research institutions. High cost was the biggest complaint about Fable 5 and likely contributed to low adoption among enterprise customers. Fable 5.1 addresses this directly, costing about 25 percent less for typical workloads and roughly 45 percent less for heavily agentic tasks, with cache reads slashed from $1 to $0.25 per million tokens. The new model also posts major benchmark gains, such as scoring 52.6 percent on Terminal-Bench-Science 0.1, more than double Fable 5's 24.7 percent.

   Fable 5.1 is available immediately on all platforms, including AWS, Google Cloud, and Microsoft Azure, with developers accessing it via the API using claude-fable-5-1. Anthropic is rolling out Enterprise Frontier Safeguards (EFS) for enterprise customers, storing customer data solely on the customer's own cloud infrastructure. The company also plans to expand access to Mythos 5.1 and the detection API over time.

2. 2

   John Ternus takes over as Apple CEO, faces AI era

   John Ternus takes over as CEO of Apple today, stepping into the role as the company confronts the AI era. Apple's App Store, long a walled garden, is under threat from vibe coding enabled by OpenAI's Codex and Anthropic's Cowork, which Apple cannot ban or control. This shift may push Apple toward hardware, but competition from Jony Ive's AI device and slimmer margins loom.

   Apple's best bet may be to build a device for the AI era akin to the iPhone's role in mobile, but Jony Ive is building a contender at OpenAI, and profit margins are likely to be smaller.

3. 3

   Top AI open source projects shutting off PRs

   Top AI-native open source projects like Flue and tldraw are refusing external pull requests, often because they are AI-generated. Maintainers prefer using their own agents to create and manage PRs. Vercel's AI SDK, with over 20 million npm downloads per week, deployed a 'software factory' of agents to clear a backlog of over 1,000 open issues and almost 800 pull requests by late June. Four weeks later, Vercel says the factory authors 25–35% of merged PRs and closes 70–80% of issues.

   Astro creator Fred Schott says the 'auto-triage' system shifted control in the last six months and led to his new framework Flue. Flue and tldraw now convert every external PR into an issue or discussion rather than accept the code.

4. 4

   VS Code's 'Rubber Duck' feature lets developers get a second AI opinion

   Visual Studio Code 1.135 now includes an experimental 'Rubber Duck' feature that lets developers request a second opinion from a different AI agent, separate from the main coding assistant. It was first introduced in GitHub Copilot CLI in April 2026 and has now been adapted for VS Code. The feature addresses a key limitation of AI-assisted development: when a single AI model reviews its own work, it tends to use the same biases and may miss errors. By using a different AI model for review, developers can catch mistakes earlier, reducing the need to redo planning or coding—saving time and effort.

   The feature is available via the '/rubber-duck' command in the Copilot Agent Host session within VS Code. It is part of a series of updates in VS Code 1.132 to 1.135, which also include a hybrid Markdown editor, a comment feature for web pages in the integrated browser, and automatic hot reload for local HTML files.

5. 5

   Claude Code to permanently raise usage limits by 25%

   Anthropic is making a permanent 25% increase to the usage limits in Claude Code, effective after September 14. This replaces a temporary 50% extra-usage promo that ends in two weeks. Users get more usage than the original 100% baseline, but less than the current 150% promo level. The change follows complaints about the marketing of the '5x' and '20x' plans, which are actually about 3.5x and 6-8x multiples on the 5-hour limit.

   Starting after September 14, usage limits will be 125 units, down from 150 units during the promo but up from the 100 units before May 13. The exact date of change is September 14.

6. 6

   OpenClaw 2.0 launches, targeting enterprise AI teams

   OpenClaw creator Peter Steinberger and co-developers announced OpenClaw 2.0 over the weekend, describing it as the most significant update to the harness and platform yet. The update shifts OpenClaw from a personal agent harness to one designed for teams and enterprise workflows, with a rebuilt browser interface, shared cloud sessions, multi-user collaboration, and an expanded security model.

   The announcement comes after interest in OpenClaw cooled from its peak in March 2026, making this release a key test of whether enterprises will adopt it.