Microsoft unveils Visual Studio Weekly: Copilot Memories, AI testing, and custom agents

Microsoft’s latest IDE upgrade is more than a cosmetic refresh; it embeds a persistent “memory” layer into GitHub Copilot that records developer preferences and propagates them across sessions. According to Hector Flores, the new Copilot‑memories feature lets users capture conventions—such as preferring Dapper over Entity Framework or enforcing a team’s async/await style—either at the personal level (saved to %USERPROFILE%/copilot‑instructions.md) or the repository level (via .github/copilot‑instructions.md). The IDE then automatically applies those rules in subsequent prompts, effectively turning ad‑hoc corrections into lasting institutional knowledge. Flores notes that a new “/generateInstructions” command can bootstrap a starter instruction file by scanning existing code, eliminating the “blank‑page syndrome” that has plagued AI‑assisted coding tools.

The second headline feature, GA‑level AI‑generated .NET unit tests, moves Copilot from a code‑completion aide to a full‑cycle testing agent. As Flores explains, developers invoke the agent with the “@Test” command in Copilot Chat, specifying scope—from a single method to an entire solution—and the tool produces, builds, runs, and iteratively refines tests until a stable baseline is reached. The workflow integrates with Visual Studio’s Test Explorer, reports coverage changes, and links directly to the generated files, thereby closing the loop that traditional code‑generation models leave open. By handling compilation failures and automatically fixing them, the feature promises to shave hours off the manual test‑writing process, a claim that aligns with Microsoft’s broader strategy to embed AI deeper into the software development lifecycle.

Beyond the headline capabilities, the release bundles a suite of quality‑of‑life enhancements that cumulatively boost developer productivity. Flores highlights “syntactic line compression,” which frees roughly 25 % of vertical screen real estate, as well as faster scrolling via Alt + scroll‑wheel and middle‑click scrolling. While these tweaks may appear modest in isolation, they address the ergonomic friction points that accrue over long coding sessions, reinforcing Microsoft’s message that the update is “shipping features that change how you work” rather than merely padding the changelog.

Analysts see the move as a defensive play against a rapidly diversifying AI‑assistance market. Competitors such as Google’s Gemini‑based tools and open‑source Copilot alternatives have begun offering context‑aware suggestions, but none currently combine persistent instruction sets with an end‑to‑end testing loop inside a mainstream IDE. By locking these capabilities into Visual Studio 2026—a platform with a reported 70 % market share among professional .NET developers—Microsoft is positioning its AI stack as the default productivity layer for enterprise development teams, a strategy that mirrors its broader AI‑first push across Azure and Office.

The timing of Visual Studio 18.3 also dovetails with Microsoft’s larger AI outreach, including its “Microsoft Elevate” program that funds AI training and research in schools, as noted in Bloomberg’s coverage of the company’s educational initiatives. While the Bloomberg pieces focus on safety and classroom transformation, they underscore Microsoft’s intent to embed AI responsibly across its ecosystem. In the developer arena, the new memory and testing features represent a concrete step toward that vision: an IDE that not only assists in writing code but also learns, enforces, and validates the standards that organizations rely on to ship reliable software at scale.