Key Signals

• Copilot CLI v1.0.11 ships full monorepo support and organizational MCP policy enforcement. Custom instructions, MCP servers, skills, and agents are now discovered at every directory level from the working directory up to the git root — a significant improvement for teams with monorepo codebases. Organization-level policy for third-party MCP servers is now enforced for all users, and a new ~/.agents/skills/ directory aligns personal skill discovery with VS Code's GHCP4A extension default. Background agent progress also now surfaces in read_agent and task timeout responses, improving visibility into long-running agentic workflows. [1]

• Gemini 3.1 Pro is now available across all major GitHub Copilot surfaces, including JetBrains IDEs, Xcode, and Eclipse. The model is accessible in public preview through the chat model picker in agent, ask, and edit modes for Copilot Enterprise, Business, Pro, and Pro+ tiers. Business and Enterprise administrators must explicitly opt in via the Gemini 3.1 Pro policy in Copilot settings. This broadens Copilot's multi-model strategy beyond its existing lineup and gives developers access to Google's frontier model directly within their IDE workflows. [2]

• OpenCode v1.3.0 delivers GitLab Agent Platform integration, git-backed session review, and Node.js runtime support. GitLab Agent Platform is now fully supported with automatic discovery of workflow models from GitLab instances, enabling workflow models to use OpenCode's local tools over WebSocket. The release also adds git-backed session review modes that let users review uncommitted changes and branch diffs directly within OpenCode, and multistep authentication now works with providers like GitHub Copilot for Enterprise. Node.js is now a supported runtime alongside Bun, expanding the project's reach significantly. [3]

• Typed languages are winning the AI coding era, with TypeScript overtaking both Python and JavaScript as GitHub's most-used language. A 2025 academic study found that 94% of LLM-generated compilation errors were type-check failures, and TypeScript grew by over 1 million contributors in 2025 (+66% YoY) to an estimated 2.6 million developers. GitHub's Andrea Griffiths argues the shift "isn't a new language — it's a shift in which existing languages win," as strongly typed languages impose clearer constraints on AI, producing more reliable code. Even shell scripting in AI-generated projects leaped by 206% as AI absorbed the friction that made it painful. [4]

• Enterprise MCP strategy is crystallizing around a hybrid API/MCP approach rather than a wholesale replacement. APIs remain essential for controlled, deterministic access to sensitive data requiring complex authorization, while MCP enables dynamic tool discovery by AI agents without documentation overhead or additional token costs. Wrapping existing APIs with MCP using Spring AI's tool command is emerging as a practical bridge strategy, though each API needs individual analysis. The key tradeoff: APIs are static and inflexible for agentic workflows, but MCP's non-deterministic nature requires governance through an MCP Gateway. [5]

• Meta is aggressively deploying internal AI agents, with Zuckerberg himself building a personal AI CEO agent. Employees across Meta are using personal agent tools like "My Claw" and "Second Brain". AI tool usage is now a factor in performance reviews, and Meta has established a new applied AI engineering org with ultra-flat structures of up to 50 ICs per manager. The company also acquired Manus and Moltbook, signaling that agentic tool adoption is accelerating in enterprise. [6]

AI Coding News

• A comprehensive 12-hour Claude Code Essentials course was published on freeCodeCamp, covering agentic workflows from beginner to advanced. Developed by Andrew Brown, the course teaches the "Agentic Loop" concept — how the AI observes a problem, thinks of a solution, and uses tools to execute the plan. It covers installation across Windows, Linux, and Mac, session management, and API cost monitoring across providers including AWS Bedrock and Google Vertex AI. The course reflects the maturing Claude Code ecosystem and growing demand for structured education around terminal-based agentic coding tools. [7]

• A practical tutorial demonstrates building an AI-powered Docker container monitoring agent using Claude for real-time diagnosis and auto-remediation. The "Container Doctor" project uses the Docker API to poll container logs every 10 seconds, detect error patterns, send structured prompts to Claude for JSON-formatted diagnosis, and optionally auto-restart unhealthy containers with conservative rate limiting. The architecture outsources all reasoning to Claude while the Python agent handles plumbing, representing a growing pattern of using LLMs as the "brain" in DevOps automation with guard rails like 3-restart-per-hour limits and Slack alerting. [8]

• Prompt engineering patterns are converging around four essential developer techniques: few-shot, chain-of-thought, role, and tool-augmented prompting. LLMs follow an instruction hierarchy where system instructions carry the most weight, followed by developer instructions, then user input. The key insight is that engineering a good prompt reduces variance and increases reliability — strong prompts require clear instructions, appropriate context, explicit constraints, and specified output formats. Tool-augmented prompting, which gives models access to real functions like CI/CD status queries, is becoming especially relevant for developer workflows. [9]

• Walmart abandoned OpenAI's Instant Checkout after in-ChatGPT purchases converted at only one-third the rate of traditional website checkouts. Daniel Danker, Walmart's EVP of product and design, called the experience "unsatisfying." OpenAI is phasing out Instant Checkout entirely in favor of app-based merchant checkout, and Walmart will embed its own chatbot Sparky inside ChatGPT instead. A similar merchant-controlled integration is coming to Google Gemini next month, suggesting that AI-mediated commerce needs to stay closer to existing merchant platforms rather than being handled inside the AI interface. [10]

Feature Update

• GitHub Copilot CLI v1.0.11 adds monorepo support, MCP policy enforcement, and refined session management. Custom instructions, MCP servers, skills, and agents are now discovered at every directory level up to the git root. Organization policy for third-party MCP servers is enforced for all users with a warning when servers are blocked by allowlist policy. The /clear command now fully abandons the current session while /new starts a fresh conversation keeping the old session backgrounded, and both commands accept an optional prompt to start the new session. Additional improvements include ~/.agents/skills/ for personal skill discovery, extension hook merging from multiple extensions, sessionStart hook context injection, per-session working directories with /cd, MCP OAuth compatibility with non-standard authorization metadata URLs, and background agent progress surfacing in read_agent responses. [1]

• GitHub Copilot expands Gemini 3.1 Pro access to JetBrains IDEs, Xcode, and Eclipse in public preview. The model is available for Copilot Enterprise, Business, Pro, and Pro+ users through the chat model picker in agent, ask, and edit modes across github.com, GitHub Mobile, VS Code, Visual Studio, JetBrains IDEs, Xcode, and Eclipse. Administrators for Business and Enterprise organizations must enable the Gemini 3.1 Pro policy in Copilot settings. [2]

• OpenCode v1.3.0 ships GitLab Agent Platform support, git-backed session review, Node.js runtime, and xAI Responses API integration. GitLab Agent Platform enables automatic workflow model discovery with local tool access over WebSocket. Git-backed session review adds modes for reviewing uncommitted changes and branch diffs directly within OpenCode. Multistep authentication now works with enterprise providers including GitHub Copilot. The xAI Responses API integration improves reasoning model performance in long multi-turn conversations with encrypted content support. The release also includes 30+ fixes across desktop app, terminal, providers, and authentication, with contributions from 16 community members. [3]

• Gemini CLI ships v0.36.0 nightly and three v0.35.0 preview patch releases. The nightly build includes 12 changes: onboarding telemetry setup, BeforeTool hook 'ask' decision support, browser session mode warnings, global session and persistent approval for web_fetch, plan mode state transition fix to prevent freezes, and skill activation tool call recording in chat history. The v0.35.0-preview.3 through v0.35.0-preview.5 releases are cherry-pick patches to the stable preview branch. [11][12]

• OpenAI Codex CLI publishes two Rust-based alpha builds (v0.117.0-alpha.9 and v0.117.0-alpha.10). These are incremental development builds with no detailed changelogs provided, continuing the Rust rewrite of the Codex CLI tool. [13]