Microsoft Research Unveils New AI-Powered Development and Analysis Tools

개발 도구 | Wed Jun 17 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 6 sources

Developer-focused AI tooling updates include Microsoft's autonomous malware analysis agent Project Ire, the Data Formulator 0.7 enterprise data analysis tool, and Browser Use Cloud's Firecracker-based infrastructure rebuild.

Analysis

[Microsoft Research] disclosed Project Ire's autonomous identification of a LOTUSLITE variant ^[1]

• LLM-based autonomous malware classification agent
• Generates function-level behavioral reports
• Detects variants based on TTPs without relying on IOCs
• Automatically invokes decompiler and binary analysis tools

[Microsoft Research] released Data Formulator 0.7 open-source AI data analysis tool ^[2]

• Integrates DBs
• warehouses
• BI
• and object stores via Data Connectors
• Context-aware agents support data preparation
• exploration
• and visualization
• Provides an interactive multimodal interface without requiring SQL or coding
• Manages branching analysis workflows

[Browser Use Cloud] rebuilt its cloud browser infrastructure on Firecracker atop EC2 ^[3]

• Sub-second session startup at $0.02 per hour for reduced cost
• Runs Firecracker inside nested VMs on EC2
• Discarded the previous Unikraft unikernel approach
• Implements autoscaling for traffic spikes

[IETF] published RFC 10008 standardizing the HTTP QUERY method ^[4]

• Defines a safe and idempotent query-only method
• Addresses GET URI length limits and POST semantic ambiguity
• Supports caching and automatic retries

[Hugging Face] published a PyTorch profiling series with optimization guidance from nn.Linear to fused MLPs ^[6]

• Compares profiles of nn.Linear vs. matmul-add operations
• Analyzes PyTorch dispatch chains including aten::t
• Experiments run on an NVIDIA A100-SXM4-80GB
• Explains the internals of torch.compile

[SIGARCH] argued for a return to full-system timing simulation ^[5]

• Accurately reflects interactions among many-core CPUs
• accelerators
• and the OS
• Highlights system call and API dependencies of agentic AI workloads
• Revisits methodologies based on gem5
• QEMU
• and Simics
• Proposes statistical sampling to reduce simulation cost

Sources

• [1] Ire identifies another LOTUSLITE specimen - Microsoft Research Blog
• [2] Data Formulator 0.7: AI-powered data analytics for enterprise data - Microsoft Research Blog
• [3] How we run Firecracker VMs inside EC2 and start browsers in less than 1s - Hacker News
• [4] RFC 10008: The new HTTP Query Method - Hacker News
• [5] The Return of Rigorous Full-System Timing Simulation - Hacker News
• [6] Profiling in PyTorch (Part 2): From nn.Linear to a Fused MLP - Hugging Face Blog