GitHub Copilot has introduced updates to its integration with Visual Studio Code that aim to improve the efficiency of coding sessions by enhancing context handling and model routing. These changes are designed to allow Copilot to work smarter by reusing information, thereby reducing the need for repeated inputs by developers.
The updates focus on increasing prompt caching and on-demand loading of tool definitions. This means that recurring information such as instructions, repository context, and available tools are efficiently prepared and stored during longer coding sessions.
This improvement facilitates smoother operations by limiting the redundancies that Copilot must process, thereby enhancing resource use.
GitHub Copilot now automatically selects the most appropriate model for a given task. This allows the platform to tailor its capabilities to the complexity of tasks such as making quick edits or implementing multi-file changes, without overwhelming users with choices.
This strategic selection improves the Copilot's performance, ensuring the right balance between speed and task nature.
By continuously evaluating the GitHub Copilot agentic harness through benchmarks, GitHub ensures efficiency and performance are in line with industry standards, showing improvements similar to competitors but with lower token consumption.
In addition, rewrites of the Copilot code review instructions have resulted in a 20% reduction in review costs, enhancing overall efficiency without compromising quality.
These improvements in context handling and model selection for GitHub Copilot indicate an optimized workflow for developers, with GitHub aiming to enhance productivity by reducing resource wastage and streamlining operations.
Such updates are significant as they could influence workflows across various software engineering environments, improving both individual and organizational coding efficiency.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's briefNew every morning, and the back catalogue is archived by date.
Public contributions on GitHub more than doubled to 2.5 billion in 2026, indicating continued growth in open source development. GitHub released two new open source projects: Replica, a design harness, and the Open Source Accessibility Framework.
GitHub has developed the Agent Host Protocol (AHP), an open standard enabling AI agent sessions to persist and transfer across different devices and applications, including VS Code, GitHub Copilot app, and mobile apps. This protocol addresses the limitation of agents being tied to single application instances, allowing developers to maintain agent state and history across various clients without custom integrations.
GitHub re-engineered the diff surface in its Copilot app to handle pull requests with over a million changed lines and thousands of files without performance degradation. This change addresses the bottleneck created by AI coding agents generating large pull requests, shifting the challenge from writing code to efficiently reviewing it. The new design separates deterministic code height from dynamic comment blocks to improve rendering performance.
GitHub has introduced "agent skills," a method for capturing and reusing repeatable instructions for AI agents. These skills are versioned directories centered on a SKILL.md file, defining a task's purpose, inputs, and outputs. This allows for consistent execution of tasks across different agents, such as verifying AI-generated content.
GitHub has made Agentic Workflows generally available, allowing coding agents to perform repository tasks within GitHub Actions. This release includes isolation mechanisms, a threat-detection job, and budget controls to manage AI credit spending.
Google has integrated CodeMender, an autonomous code-security agent, directly into Git pull requests via Google Cloud Run. This tool identifies and automatically proposes fixes for software flaws within the developer's existing workflow, aiming to reduce the time developers spend on vulnerability remediation.
This guide outlines how developers can integrate AI agents, specifically GitHub Copilot, into their software development workflows. It details how agents can assist across planning, building, verifying, and observing stages to improve efficiency. The article suggests using Copilot for consolidating information, generating actionable tasks, and automating routine triage.
GitHub announced a new Git storage architecture designed to handle thousands of writes per second and scale reads on demand. This architecture supports 5,000 concurrent agents and 7,200 pushes per second, providing a foundation for agentic development for hundreds of millions of developers. The system minimizes coordination and decouples storage from compute, allowing independent scaling of Git infrastructure components.
GitHub Copilot introduced an upgrade agent to help .NET applications migrate from Windows x64/x86 to Windows Arm64. This agent assesses projects, identifies migration blockers, creates a plan, and updates code, dependencies, and configuration. The tool aims to reduce the manual effort involved in porting applications to Arm64 architecture.
GitHub launched Copilot Fusion, a new system for selecting and orchestrating AI models for coding tasks, at GitHub Universe. This system combines Project HydraFusion and Copilot Auto Model Selection, aiming to reduce developer effort in choosing appropriate models. Copilot Fusion reportedly matches Claude Code with Opus 5.5 Max performance at half the API cost and 40% less time per task.
GitHub announced new products at Universe 2026, including the GitHub Copilot app and Copilot Fusion, to support an agentic software delivery system. These tools aim to integrate AI assistants and agents across the development lifecycle, moving from idea to completed software more efficiently. The focus is on enabling developers to orchestrate work and leverage AI for tasks beyond code generation, while maintaining control over the process.
GitHub's Octoverse report indicates developers pushed code 17.9 billion times in the past 12 months, exceeding 2024 and 2025 combined. This growth is attributed to the increased use of AI agents, with 40% of pull requests now initiated with an agent.
Entire, a startup founded by former GitHub CEO Thomas Dohmke, launched its GitHub alternative featuring a new agentic code review system called Trails. This platform aims to address code review bottlenecks in an era of AI-generated code by replacing traditional pull requests with agent-scored evaluations.
GitHub's Bug Bounty program highlighted researcher @r31n for consistently uncovering high-impact vulnerabilities in GitHub Enterprise Server. This recognition emphasizes the program's reliance on external security researchers to enhance platform security and the value of detailed vulnerability reports.
Three Java Champions with extensive experience shared their methods for using GitHub Copilot to enhance productivity in Java development. They provided examples of how the AI tool assisted in tasks like web application development and code analysis, significantly reducing implementation time. The developers emphasized the importance of providing context, validating AI-generated code, and maintaining accountability for shipped software.
The GitHub Copilot SDK now allows developers to use their own model deployments for inference, rather than relying solely on GitHub's default models. This functionality is enabled through ProviderConfig, which routes model inference to user-controlled providers like OpenAI, Anthropic, Ollama, Microsoft Foundry Local, or any OpenAI-compatible endpoint.
AI is shifting software development from command-based interactions to an intent-driven workflow, where systems understand context and execute tasks autonomously. This change allows developers to focus on defining outcomes and reviewing results, rather than tool navigation and manual coding.
GitHub developed a new context-aware secret classifier to improve the accuracy and efficiency of its secret scanning. This classifier reduces false positives and enables generic secret detection for push protection, preventing credentials from entering repositories.
GitHub Copilot can be configured to understand specific codebase conventions through custom instructions and agent skills. These features allow developers to provide context on architecture, naming, and repeatable procedures, improving Copilot's relevance to project-specific tasks.
GitHub migrated the runtime for Copilot CLI, app, and SDK from TypeScript/Node.js to Rust, replacing over 800,000 lines of code using AI assistance. This migration reduced client startup and session creation times from 5.25 seconds to 292 milliseconds and allows direct embedding into host applications.
GitHub Copilot will soon automatically route coding tasks to run either locally or in cloud models, with this functionality expected by the end of October. Microsoft has not clarified what repository context is sent to cloud models, raising data privacy concerns for developers and teams with strict data-handling policies.
GitHub highlighted @vaib25vicky, a top bug bounty researcher specializing in authorization and access control, as part of Cybersecurity Awareness Month. The article details their approach to finding vulnerabilities and GitHub's restructured bug bounty program, which rewards high-quality submissions and includes a VIP program.
GitHub is rebuilding its Git infrastructure to support agentic software development, which involves concurrent work by developers and AI agents. This initiative addresses the increased demand for write throughput and faster commit turnaround driven by agent activity, with Git activity on GitHub more than doubling in the past year.
GitHub and Microsoft introduced ReviewBench, an open benchmark to evaluate AI code review agents' ability to identify issues in pull requests. GitHub Copilot code review currently ranks highest on this new benchmark, which uses 219 public pull requests across 19 languages for evaluation. This benchmark provides a standardized method for comparing the performance of various AI-assisted code review tools in the industry.
GitHub has made its new dashboard experience the default for all users after a preview period. The updated dashboard consolidates active agent sessions, issues, and pull requests, and separates the activity feed into its own tab.
GitHub launched a new computer use feature for Copilot CLI and its desktop app, allowing it to interact with applications on macOS and Windows, including older GUI-only software. This enables Copilot to perform tasks like reading app content, clicking, typing, and moving information between applications. The update positions GitHub to compete with similar functionalities already offered by OpenAI's Codex and Anthropic's Claude Code.
Cloudflare announced a competition for developers to build a new Git platform on its Workers and Artifacts services. This initiative aims to explore new ways for agents to collaborate on codebases, addressing challenges posed by AI-driven development.
The GitHub Copilot app now features customizable 'canvases,' allowing users to create tailored interfaces like kanban boards or checklists using natural language prompts. This functionality enables both users and the AI agent to interact with and update the same interface simultaneously, adapting to specific workflow needs.
GitHub Security Lab developed the Fuzzing Taskflow, an autonomous AI agent that automates continuous fuzzing for C/C++ projects by identifying entrypoints, writing harnesses, running fuzzers, and triaging crashes. This tool aims to reduce the human effort required in traditional fuzzing processes, making security testing more efficient.
The GitHub Copilot app has re-architected its pull request view to handle extremely large pull requests, specifically addressing performance issues caused by numerous inline comments. This improvement allows developers to review extensive code changes and their associated discussions more efficiently, which is crucial for broad refactors and migrations that cannot be easily split into smaller changes.
GitHub rewrote its Copilot agent runtime, previously in TypeScript, into over 800,000 lines of Rust using its own Copilot coding agents. This migration highlights the increasing role of AI in large-scale code refactoring projects, significantly reducing development time and effort.
The GitHub Copilot agent runtime, which underpins various Microsoft and GitHub products, has been rewritten from TypeScript/Node.js to Rust. This migration, largely completed by AI agents, improved runtime performance significantly and allowed a single developer to accomplish a task that would typically require a team over a longer period.
GitHub has launched Project HydraFusion, a research preview for Copilot that uses runtime model orchestration to combine multiple AI models for coding tasks. This system dynamically routes requests across single, cascade, or critique execution patterns to optimize for performance and cost. HydraFusion demonstrated improved task quality and reduced estimated costs in offline evaluations.
The GitHub Copilot app now includes integrated diff, terminal, and browser panels, allowing users to review, run, and preview code changes within a single interface. This integration aims to simplify the workflow for developers interacting with AI-generated code, reducing the need to switch between different applications.
Microsoft has made GitHub Copilot code review generally available for Azure Repos, removing the previous sign-up requirement. This feature allows Azure DevOps customers to use AI-powered code review directly within Azure Repos, addressing the needs of organizations that cannot easily migrate to GitHub.
A research team outlined its methodology for evaluating AI coding agents like Claude, Codex, and Cursor, involving 17,000 experimental runs. The approach uses a panel of 75 simulated repositories and various engineer personas to test agent performance on real-world coding tasks.
The GitHub Copilot app now allows users to run multiple AI agent sessions in parallel, each operating independently on its own Git worktree. This functionality enables developers to execute several tasks concurrently without interference, improving workflow efficiency.
GitHub Copilot has implemented four changes to its AI coding agents to improve cost efficiency without sacrificing task quality. These changes focus on preserving useful context, removing valueless formatting, shortening instructions, and delivering completed background work directly, aiming to optimize for the overall task outcome rather than individual token counts.
An open source project maintainer notes a rise in AI-generated pull requests and security reports, often with AI-generated fixes, which they attribute to individuals using AI to inflate their GitHub contribution profiles. This trend is seen as a way for individuals to game the system for career advancement without genuine engagement with projects. The maintainer expresses concern that these contributions, while sometimes harmless, do not reflect true interest or effort.
The GitHub Copilot app can automate the triage of Dependabot pull requests by reviewing, grouping by risk, verifying CI status, and summarizing recommended next steps. This automation helps developers manage frequent dependency updates and identify critical changes more efficiently.
The GitHub Copilot app now includes a 'My work' pane to help users organize pull requests and issues across multiple projects. This feature provides centralized views and custom filtering options to manage ongoing and completed development tasks within the app. It matters to developers using the Copilot app by offering improved task organization and workflow management.
GitHub has launched agent apps, allowing developers to integrate services like Amplitude, Endor Labs, LaunchDarkly, and PagerDuty directly into their GitHub workflow. This integration enables developers to access insights and perform checks within pull requests, reducing the need to switch between different tools.
GitHub published a guide on how to begin using the GitHub Copilot app by writing initial prompts. The guide explains how to provide context to Copilot and iteratively refine requests in plain English, which helps new users get started with the AI tool.
AutoGPT, an open-source project, has implemented a strategy to manage the influx of AI-generated pull requests by creating an AGENTS.md file. This approach allows the project to accept contributions from AI agents while maintaining control over code quality and review processes.
GitHub describes how AI agents are changing the developer's role from solely coding to orchestrating workflows, emphasizing the need for wired, repeatable processes over one-off prompts. This shift involves designing systems for code validation, review, and shipping, with human judgment remaining for high-risk changes.
An engineer explored the internal workings of GitHub Copilot by analyzing its network traffic and memory, driven by an increase in Copilot credit usage. The investigation focuses on understanding how AI-powered desktop applications built with Electron operate at runtime.
GitHub has released the Copilot SDK for Java, a client library that allows server-side Java applications to programmatically create Copilot agent sessions, register tools, send prompts, and receive structured responses. This SDK offers a framework-agnostic and AI vendor-neutral approach for integrating AI capabilities into enterprise Java applications, supporting environments like Jakarta EE and Spring.
GitHub's legal team, composed of non-engineers, adopted GitHub Copilot CLI to build internal tools for automating repetitive tasks like contract review and drafting. This initiative enabled legal professionals to create custom solutions, such as a contract drafting tool, to improve efficiency in their daily workflows.
The GitHub Copilot app was used to modernize an outdated personal application, demonstrating its utility in tackling complex code refactoring tasks. This highlights how AI tools can assist developers in updating legacy projects that would otherwise be too time-consuming to manage manually.
This article outlines a workflow to enhance productivity with AI by focusing on the core functionalities of GitHub Copilot, rather than complex prompts or configurations. It emphasizes understanding and utilizing the underlying 'harness' of the tool for consistent results across different interfaces. The approach suggests that mastering the basic interaction model of AI tools yields greater benefits than pursuing advanced or niche techniques.
This article outlines a workflow to improve effectiveness with AI using existing features of GitHub Copilot, focusing on understanding and utilizing its core "harness." The approach emphasizes simplicity over complex prompts or configurations to achieve productivity gains.
The GitHub Copilot app provides a workspace for managing multiple AI agent sessions, allowing developers to switch between tasks without losing context. It connects agent sessions to specific projects, providing repository context for tasks like adding new components or fixing bugs. This approach aims to integrate AI assistance more deeply into the software development workflow beyond simple chat interactions.
A study analyzing the early 2026 rollout of Anthropic's Claude Code and GitHub Copilot CLI at Microsoft found significant adoption patterns and productivity increases among engineers. The findings suggest that the social network of engineers heavily influenced tool adoption, resulting in a 24% increase in merged pull requests, indicating improved engineering output during the trial period.
Copilot code review initially performed worse after adopting new shared tools. By rewriting review instructions to align with user behavior, average review costs decreased by about 20% without sacrificing quality.
The GitHub Copilot agentic harness has been evaluated for efficiency and performance through various benchmarks. Results indicate it matches task completion rates similar to competitor harnesses while consuming fewer tokens, which could optimize software development workflows.
GitHub Copilot is improving efficiency in VS Code by enhancing context handling and model routing. These updates include increased prompt caching and on-demand tool definition loading, allowing Copilot to reuse information and choose the best model for specific tasks without developer input.