GitHub expands Copilot with desktop app and canvas, bets on agent-native development

At the Microsoft Build 2025 conference, GitHub announced a significant expansion of GitHub Copilot beyond the traditional IDE, introducing a dedicated desktop application and a collaborative workspace called canvas. This move transforms Copilot from a simple coding assistant into an AI agent orchestration platform covering the entire software development lifecycle. The desktop app, available for Windows, macOS, and Linux, includes canvas, an environment where teams can ideate, refine requirements, generate plans, and iterate projects alongside AI. It also incorporates features like Agent Merge, which automates the merging of tasks from multiple agents, and autonomous code reviews based on predefined standards, according to GitHub's official blog (github.blog).

What happened?

GitHub Copilot, initially launched in 2021 as a code autocomplete plugin, has evolved into a comprehensive platform. At Build 2025, the company announced that Copilot now operates as a "control center" for agent-native development. The new desktop app allows developers to interact with AI agents in stages such as ideation, planning, and review, not just code generation. According to InfoWorld, this expansion responds to the need to reduce context switching and increase efficiency in engineering teams. Phil Fersht, CEO of HFS Research, noted that these features could accelerate delivery cycles and improve overall productivity.

Why is it important?

This announcement marks a paradigm shift: Copilot is no longer a coding assistant but becomes an AI agent orchestration platform. According to Advait Patel, senior reliability engineer at Broadcom, running multiple agents in parallel with sandboxes, canvas reviews, and Agent Merge that traverses CI is closer to cloud computing than an IDE plugin. This means CIOs must reevaluate Copilot as an AI-driven software delivery platform, not just a per-seat productivity tool. However, the announcement coincides with controversy over the switch to a usage-based billing model, which took effect this week. On GitHub forums, many users accused the company of a "bait and switch," requesting refunds or canceling subscriptions, as reported by InfoWorld.

What consequences will it have?

For analysts, the pricing change was necessary to sustain the parallel agent model. Patel argued that "you can't price compute at a flat per-seat rate," justifying metered billing. These new features are expected to drive enterprise adoption of Copilot as a platform, but also increase costs for heavy users. The success metric will shift from "lines of code generated" to operational outcomes like release velocity, code quality, and defect reduction. Additionally, competition with tools like Amazon CodeWhisperer or Google Gemini Code Assist could intensify, as they offer alternative pricing models. Historically, GitHub Copilot has been a benchmark in the AI assistant market, with over 1.8 million paid subscribers as of 2024, according to GitHub data. This move could consolidate its leadership or alienate a cost-sensitive user base.

What should readers know?

• New desktop app: Available for Windows, macOS, and Linux, it offers a dedicated environment for working with AI agents throughout the development lifecycle.
• Canvas: A collaborative space allowing teams to ideate, refine requirements, generate plans, and iterate projects alongside AI.
• Agent Merge: Automates the merging of tasks from multiple agents to achieve a specific goal, reducing manual intervention.
• Autonomous code reviews: Agents can review code based on predefined standards, freeing developers for more complex tasks.
• Usage-based billing: Users now pay based on consumption, which can be more expensive for teams that use AI intensively. This model is similar to cloud services like AWS or Azure, where cost scales with usage.

"The pricing change is justified by where GitHub is headed, not where the product is today. Running multiple agents in parallel... is closer to cloud computing than an IDE plugin, and you can't price compute at a flat per-seat rate." — Advait Patel, Broadcom

In summary, GitHub Copilot is evolving into an agent orchestration platform, promising greater efficiency but also variable costs. Developers and CIOs must prepare for a model where AI not only assists but manages complete workflows, with implications for budgets, success metrics, and team dynamics. The pricing controversy underscores the tension between technological innovation and user perception of value.