Story 01

Anthropic Releases Claude Fable 5 — the Most Powerful AI Coding Model Available to the Public

On June 9, 2026, Anthropic launched Claude Fable 5, calling it the first publicly available model from its new "Mythos" class — a tier that sits above the Opus line in raw capability. Anthropic says Fable 5 is state-of-the-art on nearly all tested benchmarks, with the biggest gains showing up on long, complex tasks.

The software engineering numbers are striking. Fable 5 scored 80.3% on SWE-Bench Pro, a benchmark that tests whether a model can fix real software engineering problems end to end. For comparison, the previous Claude Opus 4.8 scored 69.2%, GPT-5.5 scored 58.6%, and Gemini 3.1 Pro scored 54.2%. Fable 5 also topped Cognition's FrontierCode evaluation, which tests not just whether the task is done, but whether the resulting code meets production quality standards.

Stripe, an early tester, gave Anthropic a concrete real-world example. Stripe used Fable 5 to complete a migration across a 50-million-line Ruby codebase in a single day. Stripe said the same job would have taken a full engineering team more than two months by hand. Anthropic described the result as Fable 5 having "compressed months of engineering into days."

Fable 5 can also run for days at a time inside an agentic harness like Claude Code, planning its own approach, checking its progress, and refining its output as it goes. Anthropic notes the model also works with 1 million tokens of context, which means it can hold large portions of a real codebase in mind while it works.

One important note: access was suspended on June 12 due to a technical issue. Anthropic announced the suspension and said it is working to restore access. The model remains available through the Claude API, Amazon Bedrock, and GitHub Copilot. Pricing is set at $10 per million input tokens and $50 per million output tokens.

Source: Anthropic — Claude Fable 5 and Claude Mythos 5 (June 9, 2026)

Story 02

GitHub Agentic Workflows Moves to Public Preview — Write Automation in Plain English, Not YAML

On June 11, 2026, GitHub announced that GitHub Agentic Workflows has moved from technical preview to public preview. The feature lets development teams automate reasoning-based tasks — things like issue triage, CI failure analysis, and documentation updates — by writing instructions in plain Markdown rather than YAML.

Here is how it works. A developer writes a workflow description in a Markdown file and places it in the repository. GitHub Agentic Workflows compiles that description into a standard GitHub Actions YAML file. The compiled workflow then runs inside GitHub Actions using the team's existing runner groups and policy settings. Teams can use GitHub Copilot, Claude, Gemini, or OpenAI Codex as the agent that executes the logic.

Security is built into the design from the start. Agents run with read-only permissions by default. They can only write back to GitHub through a declared "safe outputs" list that the team specifies in the workflow file. Each agent also runs inside a sandboxed container behind what GitHub calls the Agent Workflow Firewall. All proposed changes are scanned by a dedicated threat detection job before they are applied.

Marks & Spencer's CTO James Hoare described the impact in GitHub's announcement: "Our developers were losing hours every sprint to repetitive work such as triaging issues, remediating vulnerabilities, maintaining dependencies, and reviewing routine changes. With GitHub Agentic Workflows, we've built a catalogue of reusable workflows spanning security, quality, and delivery that our teams can adopt across any repository. What once required hours of engineering effort can now be completed autonomously in minutes."

GitHub's open-source agentics repository already contains more than 50 prebuilt workflows covering triage, CI diagnosis, compliance monitoring, and more.

Source: GitHub Changelog — GitHub Agentic Workflows in Public Preview (June 11, 2026)

Story 03

GitHub Copilot Code Review Gets Deeper Controls — and Drops the Character Limit on Custom Instructions

On June 12, 2026, GitHub released several new controls for Copilot code review, the feature that uses an agentic AI to review pull requests inside GitHub Actions. Three changes stand out for teams using Copilot at the organization level.

First, runner configuration can now be set at the organization level. This means an administrator can pick one runner type — standard GitHub-hosted, self-hosted, or a larger runner — and that choice applies automatically to every repository in the organization. Before this change, each repository had to be configured individually. Teams can also lock the organization-level setting so individual repositories cannot override it.

Second, Copilot code review now respects content exclusion settings. Repository administrators can tell Copilot to skip specific files or directories during review using path-based rules. This addresses a common need for teams that have files containing sensitive data, generated code, or third-party code that should not be analyzed by an external model.

Third, the 4,000-character limit on custom instructions files has been removed. Teams write their coding standards and review preferences in a copilot-instructions.md file in the .github directory. Previously, Copilot stopped reading that file after 4,000 characters. That ceiling is now gone, which means teams can write more complete and detailed instructions without having to abbreviate their standards.

Source: GitHub Changelog — Copilot Code Review: New Configurations and Controls (June 12, 2026)

Story 04

Apple's WWDC 2026 Gives iOS Developers a Single API to Swap AI Models Without Rewriting Their Code

Apple held WWDC 2026 from June 8 through June 12. The biggest developer-side announcement was a new LanguageModel protocol built into Apple's Foundation Models framework. The protocol lets iOS, macOS, and iPadOS apps switch between different AI models — Apple's own on-device model, Google's Gemini, or Anthropic's Claude — by updating a Swift Package Manager dependency. No changes are needed in the application's session logic or other code.

Google confirmed that cloud-hosted Gemini models plug into the protocol through the Firebase Apple SDK. Anthropic published a Swift package implementing the same protocol. This means an app can prototype using Apple's on-device Neural Engine model and then route heavier queries to a cloud model without the developer changing anything in their app logic.

The framework also received free Private Cloud Compute access for developers with fewer than two million first-time App Store downloads. That removes infrastructure cost as a barrier for smaller development teams who want to add AI-powered features to their apps. Apple also announced that the Foundation Models framework will go open source later this summer.

Xcode 27, also announced at WWDC 2026, ships a dual-engine coding system. It uses a local Neural Engine model for real-time Swift code suggestions and routes heavier analysis tasks to Claude, Gemini, or OpenAI's agents through a cloud layer. This keeps the fast suggestions local while offloading deeper code analysis to frontier models.

Apple also confirmed that SiriKit, the older framework for building Siri integrations, will be deprecated over a two-to-three-year period as teams migrate to the new Foundation Models architecture.

Sources: MacRumors — Apple WWDC 2026 Platforms State of the Union (June 9, 2026); TechTimes — WWDC 2026 Developer Tools (June 9, 2026)

Story 05

GitHub Copilot Code Review Now Configures Runners at the Organization Level — One Setting Covers Every Repository

GitHub also made a structural change to how Copilot code review fits into enterprise environments. Starting June 12, 2026, organization administrators can set the runner type for Copilot code review once at the organization level, and that setting applies automatically across all repositories. Administrators can also lock the setting so individual repository owners cannot change it.

This matters for larger engineering teams because Copilot code review now runs on an agentic architecture powered by GitHub Actions. That architecture gives Copilot the ability to analyze a pull request across multiple files, hold context across the review, and produce more thorough feedback than a single-pass model. The runner it uses determines how much compute it has available and where that compute runs.

Teams that run their own self-hosted runners — often required when code must stay on-premises for compliance or security reasons — can now point Copilot code review at those runners without configuring each repository individually. Teams on GitHub Enterprise Cloud can also point the review agent at large runners for heavier repositories without a per-repo setup step.

The same runner configuration applies to both Copilot code review and the Copilot cloud agent when both features are enabled. That means one setting governs multiple AI features, reducing administrative overhead for teams managing many repositories.

Source: GitHub Changelog — Copilot Code Review: New Configurations and Controls (June 12, 2026)