Agent Eng News

Monday, August 17, 2026 · Past edition · 9 stories

Retrospective edition reconstructed from later captures. Paper versions may include later updates; coverage labels describe the evidence used.

Browse published editions

Filter by topic · All topics

A coding-agent reinforcement-learning framework joins research on model optimization, clinical risk and malicious agent skills.

Lead stories
Research & academia
LEGO-RL: Harness-Native Reinforcement Learning for Coding Agents
LEGO-RL targets the mismatch between native coding-agent harnesses and policy-gradient training, aiming to make rollouts trainable and reliable without changing harness control flow.
Why it matters: The approach is designed to let coding agents train within their native harnesses while addressing unreliable reward signals and rollout-policy mismatch. The reported SWE-bench Verified scores improve across all three evaluated harnesses.
Original · arxiv (new tab)Read
Research & academia
Developing Nemotron 3.5 Lightning NVFP4 with QAD Using NVIDIA Model Optimizer

Also on the radar 2

Additional source headlines. These have not received a full selected-story review.

Benchmarking the Benchmarks: Evaluating Automated Safety Benchmarks for Small Language Models (opens original in a new tab)
arxiv.org · Discovery link; source not captured
HarnessEval-W: Agentifying the Evaluation of Visual Worlds (opens original in a new tab)
arxiv.org · Discovery link; source not captured

Keep notes on the stories you read

Open each story in the reader. Summaries are shown when available; some sources provide an abstract only. With an account you can keep your own notes.

Personal notes on the sources you read, tied to your account
A Library to come back to, with export to Markdown, Notion or Obsidian
AI questions and rewrites on accounts with AI tools enabled

How Reading + AI works

Agent Eng News

Lead stories

LEGO-RL: Harness-Native Reinforcement Learning for Coding Agents

Developing Nemotron 3.5 Lightning NVFP4 with QAD Using NVIDIA Model Optimizer

Keep notes on the stories you read

Seeing beyond BMI: Estimating cardiometabolic risk with smartphone imagery

Safety, alignment & governance

CompoSkill: Compositional Skill Chain Attacks from Individually Scanner-Passing LLM Agent Skills

Infrastructure & deployment

What Does Context Compression Cost an Agent? Interaction Costs Unrevealed by Task-Completion Metrics

Agent frameworks & orchestration

The Working Set of a Coding Agent: Coherence Debt in Repository-Scale Tasks

When Agents Coordinate: Measuring Coordination in Multi-Agent AI Coding

Research & academia

Zetta ζ: An Efficient Closed-Loop Embodied Harness for Self-Evolving Physical Intelligence

ASI-Bench: At the Dawn of Artificial Superintelligence