VS Code deprecates legacy AI agent harness

// 2h agoPRODUCT UPDATE

VS Code deprecates legacy AI agent harness

Microsoft has functionally frozen Visual Studio Code's legacy Local agent harness, scheduling complete removal for February 2027. Teams must migrate extensions, permissions, and hooks to the Copilot harness to maintain background session support.

// ANALYSIS

The death of the window-bound Local harness signals VS Code's shift from an interactive editor assistant to an autonomous background agent platform. This is an enterprise security overhaul disguised as runtime cleanup.

–Stricter execution boundaries: Moving to the Copilot harness enforces centralized sandbox controls and auto-approval policies, stripping extensions of unmonitored command execution.
–Multi-session persistence: The Copilot SDK runtime decouples agents from individual editor windows, enabling features like Inbox mode and cross-project agent management.
–Enterprise settings expiration: Fallback flags like vscode.temporaryDefaultToLocalHarness will be erased in February 2027, forcing DevSecOps teams to audit permission rules now.
–Extension migration pressure: Community and internal extensions tied to legacy chat participant APIs must adopt the Copilot SDK before the freeze breaks automated workflows.

// TAGS

vs-codeai-codingcoding-agentagentidedevtoolsecurity

DISCOVERED

2h ago

2026-11-15

PUBLISHED

2h ago

2026-11-15

RELEVANCE

8/ 10

AUTHOR

OutworkVentures

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

MODEL58m ago

Grok 4.8 begins rollout to bot users

Early reports indicate xAI has begun rolling out its Grok 4.8 frontier model to select Grok Bot users following completion of pre-training and entry into reinforcement learning. The deployment tests xAI's rebuilt C/C++ inference stack designed to run massive multi-trillion-parameter architectures efficiently.

MODEL2h ago

Seedance 2.5 Flash lands on fal

ByteDance's Seedance 2.5 Flash is now accessible via fal, bringing accelerated text, image, and reference-to-video generation to the developer platform. The model supports native synchronized audio, video editing, and extensions from 4 to 30 seconds at 480p and 720p.

UPDATE2h ago

PIES Studio decouples memory from AI models

PIES Studio has introduced an external memory architecture that stores organization, application, and agent notes directly in the developer's database rather than inside vendor model silos. The decoupling lets builders swap seamlessly between local private GPUs, custom endpoints, and commercial cloud LLMs without losing application context or UI rules.