GitHub Copilot got a late-June platform bundle: MAI-Code-1-Flash is now GA for Copilot Business and Enterprise, Copilot code review now uses CLI/SDK file tools and says that trims review cost by about 20%, GitHub Desktop 3.6 adds worktrees plus Copilot-assisted merge conflict resolution, and enterprise-managed settings now support strictKnownMarketplaces in VS Code and Copilot CLI. Sources: GitHub MAI-Code-1-Flash | Copilot code review | GitHub Desktop releases | strictKnownMarketplaces | HN discussion | Reddit discussion
GitHub Copilot for JetBrains IDEs shipped a meaningful agent UX update: organization and enterprise agents are now available in the IDE, Copilot CLI can queue or steer messages while a request is running, the Agent Debug panel got a logs summary view, Claude as an agent provider is in public preview, and cloud agent is now GA. Source: GitHub Changelog
Microsoft Agent Framework 1.11.1 is a real platform step, not just a version bump: .NET now makes AgentSkillsProvider tools require approval by default, adds AOT-safe declarative workflow checkpointing, fixes checkpoint resume after version upgrades, and adds Telegram channel hosting plus Foundry environment variables in Python. The repo is still at about 11.7k stars, 2k forks, and 97 releases. Sources: release | repo
OpenAI published a new agent-economics readout, “How agents are transforming work,” alongside a Codex evidence PDF showing that 60.3% of Codex turns and 21.9% of ChatGPT turns in the sampled week used at least one external tool. The message is the same one everyone else keeps converging on: capable agents are pushing work toward longer, more cross-functional tasks. Sources: article | PDF
Anthropic’s June 26 Economic Index report is one of the better fresh telemetry drops this week: it shifts to hourly/day-level sampling, shows weekend usage moving toward AI agent design, backend architecture, API debugging, and data storage, and argues that more compute correlates with more valuable artifacts. HN discussion is already leaning skeptical but engaged. Sources: report | HN
DeepMind published “Securing the future of AI agents,” framing agent security in three layers: individual agents, multi-agent systems, and defender tooling/ecosystem resilience. This is the kind of paper that matters because it is trying to define the problem space before the incidents pile up. Sources: DeepMind blog | HN
Hong Gao’s LinkedIn post adds useful technical context to Microsoft’s agent observability push: unified Microsoft OTel Distro across Azure Monitor, Azure AI Foundry, and Microsoft 365 Copilot, AI-assisted auto-instrumentation that can get a .NET agent from zero observability to full traces in under a minute, and an Agent Fleet View for errors/latency/cost. Sources: LinkedIn | Microsoft blog
agentgateway v1.3.0 is a strong infra signal: the release adds a purpose-built UI, AI cost analysis, virtual models, reusable providers and guardrails, 13 new LLM providers, and more MCP support. The repo is already at about 3.6k stars and shows active 2026 churn. Sources: release | repo | activity