Agentic Dev Tools: Copilot API, SDK GA, and Glasswing
Curated by Dillip Chowdary • June 05, 2026 • Morning edition, IST
Today's Top Highlights
- 1.GitHub Agent Tasks API: Copilot cloud agent tasks can now be started and tracked programmatically in public preview.
- 2.Copilot SDK GA: GitHub made its agent runtime available through stable APIs across six languages.
- 3.Project Glasswing: Anthropic is expanding access to about 150 organizations across more than 15 countries.
- 4.AWS Modernization: AWS highlighted 5 to 30 minute repo scans for agentic readiness and modernization analysis.
- 5.NVIDIA AI Factories: DSX OS and Vera CPU position token operations as infrastructure work, not just model serving.
This Week in Tech
Jun 05
Audit Copilot API automations before enabling write paths.
Jun 06-07
Queue Project Glasswing-style security scans for critical repositories.
Jun 09-12
Recheck Vertex AI SDK Gemini migration timelines before June 24 removal.
GitHub: Agent Tasks REST API Enters Preview
GitHub made the Agent Tasks REST API available in public preview for Copilot Pro, Pro+, and Max users. The important shift is that teams can now start and track Copilot cloud agent work from portals, scripts, and release workflows.
- Automation Surface: Tasks can fan out migrations, prepare weekly releases, and bootstrap repositories from an internal developer portal.
- Execution Model: The cloud agent works in a separate development environment, validates code, and opens pull requests.
- Auth Paths: The preview supports classic PATs, fine-grained PATs, and OAuth tokens.
- Builder Risk: Treat agent task creation as privileged infrastructure because it can create changes across many repositories.
GitHub: Copilot SDK Is Generally Available
Copilot SDK is now generally available, giving product teams stable access to GitHub's agentic engine. The SDK exposes planning, tool invocation, file edits, streaming, and multi-turn sessions without requiring every team to build its own orchestration layer.
- Language Coverage: GitHub says the SDK is available across six languages for production integrations.
- Runtime Scope: The same agent foundation can power CI/CD assistants, internal tools, and customer-facing AI features.
- Integration Pattern: Teams should centralize policy, logging, and approval gates around the SDK before exposing write tools.
- Commercial Impact: This turns Copilot from an editor feature into an embeddable platform primitive.
GitHub: Copilot App Preview Opens Wider
The GitHub Copilot app technical preview is now available to existing Copilot Pro, Pro+, Business, and Enterprise customers. GitHub frames the app as a desktop control center where agents update a canvas and developers approve, reorder, or redirect work.
- Workflow Change: Agent management moves from chat-only loops into a visible task surface.
- Review Ergonomics: The app is designed around transcripts, diffs, approvals, and evidence of completion.
- Access Rule: Copilot Free and non-Copilot users remain on a waitlist for broader access.
- Team Signal: Multi-agent work needs a system of record, not scattered terminal sessions.
Anthropic: Project Glasswing Expands Security Access
Anthropic is expanding Project Glasswing, its coordinated effort to secure important software with frontier model-assisted code scanning. The company says initial partners have already found more than 10,000 high- or critical-severity flaws.
- Scale-Up: Access is expanding to about 150 new organizations after an initial group of roughly 50 partners.
- Geography: The new group spans more than 15 countries, with many organizations supporting critical infrastructure.
- Model Path: Anthropic references Claude Security and public frontier models such as Claude Opus 4.8.
- Governance Need: Findings need triage ownership, exploitability scoring, and patch validation before merge.
GitHub: Copilot CLI Adds Scheduling and Voice
GitHub Copilot CLI received a major refresh with generally available rubber duck and voice input features, while prompt scheduling moved into /experimental. A new experimental terminal interface adds tabs for working with issues, pull requests, and gists.
- Prompt Scheduling: /every and /after let developers run repeated or delayed prompts inside live sessions.
- Voice Input: Spoken prompts lower friction for quick terminal tasks and review loops.
- Second Opinion: Rubber duck mode gives a structured critique path before committing changes.
- Ops Pattern: Scheduled prompts still need external supervision for durable jobs and audit logging.
AWS: AI Modernization Scans Move Toward Agentic Readiness
AWS's June 1 roundup highlights AI-DLC, Claude Opus 4.8 on AWS, and modernization analysis updates. The notable builder detail is Agentic Readiness Analysis and Modernization Analysis, which scan repositories in 5 to 30 minutes per repo.
- Evidence Output: AWS describes severity-tagged findings with file-level evidence and mapped remediation guidance.
- Bedrock Link: The workflow pairs modernization analysis with tools such as Claude Code on Amazon Bedrock.
- Migration Fit: This is useful for monolith decomposition, dependency cleanup, and cloud readiness backlogs.
- Control Point: Teams should require reproducible scan artifacts before creating agent-generated pull requests.
NVIDIA: DSX OS and Vera CPU Target AI Factory Operations
NVIDIA's latest technical blog updates put the operating layer for AI factories in focus. DSX OS is described as open, modular software for operating token-producing infrastructure, while Vera CPU targets agentic workloads in large-scale AI systems.
- Operational Layer: AI factories need scheduling, observability, resiliency, and workload placement across accelerated clusters.
- CPU Role: Agentic workloads increase orchestration, data movement, and tool coordination around GPU inference.
- Cost Signal: Token throughput now depends on full-stack design, not only model choice or GPU count.
- Developer Takeaway: Build inference services with explicit queueing, routing, and trace budgets from day one.
Developer Resources
Key Takeaways
1Agent task APIs should be governed like deployment systems.
2SDK-based agents need central policy, logging, and permission boundaries.
3Security scanners are now finding flaws at volumes that require triage automation.
4Modernization scans are becoming an input to agentic code migrations.
5AI factory software is now part of developer productivity, not just data center operations.