xAI released 'Grok Build,' a coding-assistant agent. The composition leans academic—four arXiv papers plus one ITmedia report—research and press rather than an official announcement, a release with technical background. Beyond code generation, the focus is agentic development support that autonomously runs plan, implement, and fix; the through-line is less generation quality itself than the spread across vendors of a race over agents you can hand development workflows to. It reads as xAI entering a space where OpenAI and NVIDIA are ahead. But it's right after release—actual code quality, how it compares with existing agents, and adoption in real development remain to confirm.
AI Agents × Safety & Evaluation
xAI ships Grok Build coding agent
xAI ships Grok Build coding agent
✎ Story body
▲ Official & Press
Press
xAIがコーディングエージェント「Grok Build」ベータ公開。サブエージェントを並列に実行可能など
xAI opens "Grok Build" coding agent beta, runs subagents in parallel
Academic (arxiv etc.) 16 ▾
Academic
MobileMoE: Scaling On-Device Mixture of Experts
MobileMoE: sub-billion on-device MoE LMs claim a new on-device Pareto frontier
Academic
Causal Risk Minimization for High-Dimensional Treatments
Causal risk minimization for high-dimensional treatment spaces
Academic
Learning When to Think While Listening in Large Audio-Language Models
Learnable wait-think-answer control balances latency and reasoning in audio LMs
Academic
MobileGym: A Verifiable and Highly Parallel Simulation Platform for Mobile GUI Agent Research
MobileGym launches verifiable, browser-hosted mobile GUI agent simulator.
Academic
Beyond Summaries: Structure-Aware Labeling of Code Changes with Large Language Models
Paper: LLMs label code changes via 2-stage pipeline, 84% recall achieved.
Academic
Automated Benchmark Auditing for AI Agents and Large Language Models
Auto Benchmark Audit flags critical issues across 168 LLM and agent benchmarks
Academic
Bootstrap mode frequency: best-calibrated confidence for activation oracles
Academic
Peak-Then-Collapse and the Four Interface Channels of Knowledge-Graph Tool Use
Peak-then-collapse: minimal KG tool RLVR yields four recurring failure modes
Academic
CausaLab: A Scalable Environment for Interactive Causal Discovery Toward AI Scientists
CausaLab evaluates interactive causal discovery by LLM agents in a virtual lab
Academic
STORM: Internalized Modeling for Spatial-Temporal Reasoning in Video-Language Models
STORMS internalizes spatial-temporal reasoning in video-language models
Academic
MAGIC: Multimodal Alignment & Grounding-aware Instruction Coreset for Vision-Language Models
MAGIC: training-free, forward-only coreset for multimodal instruction tuning
Academic
Output distribution decides whether NLI checkers train medical RAG agents
Academic
Neural Scalable Symbolic Search Framework for Complex Logical Queries with Multiple Free Variables
NeSyS-k: neural symbolic search for multi-variable complex KG query answering
Academic
LLM agents treat semantic noise far more disruptively than surface noise
Academic
Creative Quality Alignment: Expert Tacit Knowledge Transfer via Chain-of-Thought Fine-Tuning
Empirical study confirms Calibrated Surprise metric at engineering scale
Academic
From Latent Space to Training Data: Explainable Specialization in Minimal MLPs
Coverage regularization in minimal MLPs aids dataset reconstruction