AI Coding Live News — Real-Time 𝕏/Twitter Updates & Breakthroughs

👀50

🔥99

Anthropic Officially Releases Claude Duet Research Preview: Proactive Autonomous Workflow Agent Application

Anthropic (@claudeai) officially announced Claude Duet, a research-preview application that represents a generational leap in agentic user interfaces. Accumulating over 40K likes and 4.7M views within hours, Claude Duet shifts the AI paradigm from passive query-response chatbots to a proactive autonomous partner: it surfaces what requires user attention, autonomously executes delegated long-horizon tasks, works uninterrupted in the background while users step away, and explicitly requests confirmation before executing high-impact actions on the user's behalf.

Key Takeaways

@claudeai officially launched Claude Duet as a research preview app, rapidly amassing 40K+ likes and 4.7M views
Transcends reactive chat interfaces via a proactive architecture that continuously surfaces priorities requiring human attention
Executes multi-step background tasks asynchronously through persistent 'keeps working when you step away' autonomy
Integrates rigorous safety guardrails, mandating explicit confirmation before executing irreversible external actions

Read full story

Linked Models & Agents:

⚡ Claude Opus 5.5 ⚡ Claude Sonnet 5.5 🤖 Claude Duet 🤖 Claude 🌐 Anthropic

#Claude Duet#Anthropic#Autonomous Agent#Human-in-the-Loop#Workflow#Productivity

Hugging Face Daily Papers@HuggingFace·Collected 28d ago

👀120

🔥99

ProgressCompass: Context-Aware Embodied Progress Reward Models Drive Long-Horizon Agent Planning

As embodied robotics agents tackle long-horizon tasks across household manipulation and industrial assembly, standard terminal outcome reward models (ORMs) fail to provide actionable learning signals along multi-step trajectories. While Process Reward Models (PRMs) evaluating step-by-step task progress percentages offer a viable solution, researchers from Northwestern University and CMU demonstrate in arXiv:2609.36684 that progress reward models collapse into random noise without appropriate context. The authors introduce ProgressCompass, formalizing the essential context dependencies required for embodied progress evaluation: task objectives, initial environmental baseline states, and continuous action history. Without grounding in preceding action sequences, PRMs routinely misjudge progress direction, confusing partial object states with completed goals. ProgressCompass provides an empirical testbed and modeling architectures to guide search and policy refinement in long-horizon robotic manipulation (andyzworks.github.io/progresscompass/).

Key Takeaways

Northwestern and CMU present ProgressCompass, formalizing essential context dependencies for embodied Process Reward Models
Demonstrates that progress estimation collapses without joint access to goal specs, initial baselines, and temporal action histories
Boosts long-horizon robotic task success by 38.6% and reduces interaction overhead by 41.2% via context-aware PRM trajectory guidance

Read full story

Linked Models & Agents:

⚡ ProgressCompass⚡ Process Reward Models⚡ Embodied PRM🤖 Embodied-Agent 🤖 Progress-Evaluator 🤖 Robotic-Planner 🌐 Northwestern University 🌐 Carnegie Mellon University 🌐 Hugging Face

#ProgressCompass#Embodied AI#Process Reward Models#Robotics#Long-Horizon Planning#PRM

Hugging Face Daily Papers@HuggingFace·Collected 29d ago

👀89

🔥99

VeriHarness: Google and Cambridge Rethink Agentic Verification for Long-Horizon Tasks with Over $100K in Open Datasets

As LLM agents tackle increasingly complex, long-horizon software engineering and workspace tasks, test-time output verification without ground-truth solutions or rubrics remains a critical bottleneck. Researchers from Google Cloud AI Research and the University of Cambridge uncover that classical consensus-based heuristics fail in deep multi-step workflows: disagreement frequently reveals correct alternative implementations, whereas consensus often masks shared blind spots. Motivated by this, they introduce VeriHarness (arXiv:2610.00972), transforming standard foundation models into agentic verifiers equipped with isolated workspaces, environmental execution tools, and modular verification skills. A disagreement resolver cross-checks competing claims against live environmental evidence, while a consensus challenger scrutinizes shared assertions for overlooked edge constraints. Guided by evidence-backed revisions, VeriHarness delivers substantial gains of +6.2 points on Gemini 3.5 Flash and +6.4 points on Claude Opus 4.8 across five workspace benchmarks, backed by an open-sourced pool of 26,000 rollouts produced at an evaluation cost exceeding $100,000.

Key Takeaways

Discovers that consensus heuristics fail in long-horizon agent tasks: disagreement reveals solutions while consensus masks errors
Introduces VeriHarness with sandbox-backed Disagreement Resolvers and Consensus Challengers, gaining +6.2 pts on Gemini and +6.4 pts on Claude
Releases a landmark dataset of 26,000 long-horizon rollouts produced at an evaluation cost exceeding $100,000 under Google Research

Read full story

Linked Models & Agents:

⚡ Gemini 3.5 Flash ⚡ Claude Opus 4.8⚡ VeriHarness🤖 Agentic-Verifier 🤖 Disagreement-Resolver 🤖 Consensus-Challenger 🌐 Google Cloud AI Research 🌐 University of Cambridge 🌐 Google Research

#VeriHarness#Agent Verification#Long-Horizon#Majority Voting#Execution Sandboxes#Self-Improvement

AI Benchmarks Track@aibenchmarks·Collected 46d ago

👀58

🔥94

View on 𝕏 ↗

Android Bench 2.0 Released: Shifting Evaluation to Multi-Day Long-Horizon Tasks

Researchers released Android Bench 2.0, moving benchmarks from single-function bug fixes to complex multi-day Long-Horizon Tasks (LHT). The suite evaluates coding agents on building apps from scratch, complex multi-library version migrations, and cross-platform verification.

Key Takeaways

Deprecates trivial unit bug benchmarks in favor of multi-day full-system app synthesis
Tests real-world architectural scaffolding, breaking library upgrades, and UI regression
Top frontier agents currently achieve only 31.4% success, exposing long-horizon bottlenecks

Read full story

Linked Models & Agents:

⚡ Claude 5.1⚡ GPT-6 Astra 🤖 Cursor 🤖 Claude Code

#Android Bench 2.0#Long Horizon Tasks#Coding Agents#Benchmarks#Software Engineering

xAI@xAI·Collected 47d ago

👀62

🔥93

View on 𝕏 ↗

xAI Unveils Grok Build Memory System and Deploys NVIDIA Vera CPUs for Agents

xAI released a native Memory system for Grok Build, enabling persistent cross-session architectural and workflow context. The lab also announced the deployment of NVIDIA Vera CPU infrastructure to power next-gen autonomous Grok agents and long-horizon tasks.

Key Takeaways

Persistent memory across project threads saves coding preferences and conventions
Deploys NVIDIA Vera CPU server clusters optimized for high-concurrency agent workflows
Enhances long-horizon recall and planning stability for autonomous development

Read full story

Linked Models & Agents:

⚡ Grok 4.7 🤖 Grok Agent 🌐 xAI 🌐 NVIDIA

#xAI#Grok Build#Memory#NVIDIA Vera#Agents

Arena.ai@arena·Collected 62d ago

👀65

🔥94

View on 𝕏 ↗

LMArena: Gemini 3.8 Flash (High) +5.94% net on Agent Arena, ranks #14

LMArena (@arena) breaks down Gemini 3.8 Flash (High) vs 3.7 Flash (High) on Agent Arena: net improvement jumps from +0.84% (#32) to +5.94% (#14). Agent Arena scores relative gains on real long-horizon tasks with web, filesystem, and terminal tools. Among five signals, Praise vs Complaint improves most (+14.78% vs -1.60%); Confirmed Success and Bash Recovery also rise, while Steerability remains a relative weakness but better than 3.7.

Key Takeaways

Agent Arena net +5.94% and #14 overall, up from +0.84%/#32 on 3.7 Flash (High).
Praise vs Complaint +14.78% vs -1.60%; Confirmed Success and Bash Recovery also improve.
Steerability still weak (-1.43%) but better than 3.7; Tool Hallucination remains clean for both.

Read full story

Linked Models & Agents:

⚡ Gemini 3.8 Flash ⚡ Gemini 3.7 Flash 🤖 Cursor Composer (Agent Mode)🤖 Google Antigravity (AGY)🌐 Google DeepMind

#Gemini 3.8 Flash#Agent Arena#LMArena#Benchmark#Google DeepMind#Bash Recovery#tool use