Copilot app’s ecological ledger: cross-device sessions, billing by API and custom Agent

At the 2026 Microsoft Build Conference, the release of the GitHub Copilot app is not just a desktop experience upgrade, but a reconstruction of the entire ecosystem of the "Agent native development environment" by GitHub. Compared with "another AI programming interface", what is more worthy of dismantling is the product ledger behind it.

Move the session from "Terminal" to "Device Network"

The core product concept of Copilot app is to "embedde the working methods that developers are accustomed to". It supports cross-device session continuation: local sessions opened in VS Code or CLI can be continued on the mobile phone - officially called Remote control for GitHub Copilot sessions, it is officially available, covering github.com and GitHub Mobile. Developers' "ongoing thinking" is no longer locked in an IDE process.

Companion capabilities also include stacked sessions and PR workflows, Canvases interactive workspace, and Copilot code review optimized based on shared Unix-style code exploration tools to reduce review costs.

Agentic harness for 20+ models

Another underrated sign is model strategy: Copilot offers an Agentic harness selection of 20+ models. This means that Copilot is no longer bound to a single model, but becomes an "Agent scheduling layer" that allows teams to select the most appropriate model according to tasks. This forms a direct competition with the multi-model strategy of Cursor and Trae, and truly returns the power of model selection to developers.

Synchronous reconstruction of billing model and product layering

Actions at the ecological level are equally intensive: GitHub launched a custom Agent (covering observability, IaC, security and other fields), Copilot's billing method was changed to according to the API listed price, and the Copilot Pro+ plan was launched. From "subscription by seat" to "billing by API consumption", this essentially turns Copilot from a software product to a platform-based service - the greater the usage of the team, the greater the revenue contribution.

GitHub summed up the vision of this transformation in one sentence: from "pair programmer" to "peer programmer". From an industry perspective, this narrative pushes Copilot from "assistant in writing code" to "an AI colleague who is equal to developers", and is also the common end point of the competition for Agent's native development environment.

Several directions worth tracking in the future:

1. Realistic experience of cross-device sessions: Contextual fidelity of continuing desktop sessions on mobile.
2. The cost impact of API billing on small and medium-sized teams: How will the cost structure change from subscription to usage billing.
3. Richness of the customized Agent ecosystem: The implementation quality of partner Agents in fields such as observability, IaC, and security.
4. 20+ model selection rights vs ecological binding: How to balance multi-model openness and the Microsoft/OpenAI ecosystem.