Devlery

Blog

Notes and analysis on AI development.

Cloudflare Agents SDK 0.14 Adds Recovery for Interrupted Agent Work

Cloudflare Agents SDK 0.14 Adds Recovery for Interrupted Agent Work

Cloudflare Agents SDK 0.14 adds Agent Skills, Telegram messenger support, scheduled tasks, ThinkWorkflow, and hardened recovery for durable agents.

QVAC TurboQuant brings KV cache compression to local AI

QVAC TurboQuant brings KV cache compression to local AI

Tether QVAC published a TurboQuant implementation for local inference, with 3-bit KV cache compression, Vulkan kernels, long-context benchmarks, and caveats.

Copilot app expands with canvases for agent work

Copilot app expands with canvases for agent work

GitHub expanded the Copilot app technical preview with canvases, worktrees, browser validation, cloud sessions, and new cost questions.

Foundry Hosted Agents near GA with per-session sandboxes

Foundry Hosted Agents near GA with per-session sandboxes

Microsoft Foundry expanded its agent operations layer with hosted agents, Toolboxes, procedural memory, and Agent Optimizer.

Postman AI Engineer brings API contract checks into pull requests

Postman AI Engineer brings API contract checks into pull requests

Postman introduced AI Engineer, an API-aware agent that uses Context Graph, sandboxed execution, and PR QA to review API contracts.

Cisco Cloud Control brings Codex into network operations

Cisco Cloud Control brings Codex into network operations

Cisco Cloud Control puts AI agents, Codex, policy, audit logs, and digital-twin validation into one network operations platform.

Work IQ APIs reach GA on June 16, setting Microsoft 365 agent pricing

Work IQ APIs reach GA on June 16, setting Microsoft 365 agent pricing

Microsoft Work IQ APIs reach GA on June 16 with M365 context, 10 generic tools, Copilot Credits pricing, and ACS control points.

Copilot Usage Billing Is Live, and Code Review Now Spends Actions Minutes

Copilot Usage Billing Is Live, and Code Review Now Spends Actions Minutes

GitHub Copilot moved to AI Credits on June 1. Copilot code review now consumes both AI Credits and GitHub Actions minutes.

MAI-Thinking-1 35B Shows Microsoft Reducing Its OpenAI Dependence

MAI-Thinking-1 35B Shows Microsoft Reducing Its OpenAI Dependence

Microsoft introduced MAI-Thinking-1 and seven in-house MAI models, bringing 35B active reasoning, 256K context, and Foundry private preview into focus.

GitHub Copilot automations turn cloud agents into scheduled PR workers

GitHub Copilot automations turn cloud agents into scheduled PR workers

GitHub Copilot cloud agent automations add scheduled and repository-event triggers, with sandboxing, model policy, and AI Credits billing attached.

Anthropic Expands Project Glasswing to 150 Organizations as Mythos Turns Vulnerability Discovery Into a Patch Bottleneck

Anthropic Expands Project Glasswing to 150 Organizations as Mythos Turns Vulnerability Discovery Into a Patch Bottleneck

Anthropic expanded Project Glasswing to about 150 new organizations. The hard part is shifting from Claude Mythos findings to validated disclosures, patches, and deployments.

Microsoft MXC previews agent isolation for Windows and Copilot CLI

Microsoft MXC previews agent isolation for Windows and Copilot CLI

Microsoft used Build 2026 to frame Windows as an AI agent runtime with MXC, Windows 365 for Agents, Agent 365 identity, and Aion local models.