Daily Agentic Field Watch - 2026-07-15 21:00 UTC

• Microsoft Agent Framework for Go is the newest official expansion of MAF, adding a Go implementation, samples, and docs to the same production agent stack that already spans Python and .NET. Source: https://github.com/microsoft/agent-framework-go
• Microsoft’s ai-agents-for-beginners course was updated on 2026-07-14 to move off deprecated models, migrate remaining notebooks to the stable Microsoft Agent Framework API, and validate against a live Foundry deployment. Source: https://github.com/microsoft/ai-agents-for-beginners/blob/main/CHANGELOG.md
• GitLost is the sharpest agent-security signal this pass: Noma Security says a crafted public GitHub issue could make GitHub Agentic Workflows leak private repo data through prompt injection, and the HN discussion is already picking apart the attack path. Sources: https://noma.security/blog/gitlost-how-we-tricked-githubs-ai-agent-into-leaking-private-repos/ ; https://news.ycombinator.com/item?id=48827858
• Microsoft Research’s SciAgentArena publication benchmarks AI agents on scientific challenges across scales, pushing evaluation toward scientific reasoning and real failure modes rather than generic QA. Source: https://www.microsoft.com/en-us/research/publication/benchmarking-ai-agents-for-addressing-scientific-challenges-across-scales/
• LangGraph’s latest release train remains active: langgraph==1.2.9 and langgraph-cli==0.4.31 shipped on July 9-10, and the repo still positions LangGraph as a low-level framework for long-running, stateful agents. Sources: https://github.com/langchain-ai/langgraph/releases ; https://github.com/langchain-ai/langgraph
• Google Research’s ReasoningBank is a useful cross-vendor memory paper: it learns from both successful and failed experiences so agents can keep improving after deployment. Source: https://research.google/blog/reasoningbank-enabling-agents-to-learn-from-experience/