2026-10-16

AI for Software Development

MLExpert: Towards a Generalist Agentic Model for Autonomous Machine Learning Engineering

Relevance: Trains an agentic model to autonomously perform long-horizon machine learning engineering, including writing code, running experiments, and adapting optimization strategies from execution feedback. This matches the software development focus on AI assisting with code generation, experimentation, and iterative debugging. It is a strong fit for AI-assisted engineering workflows, though its emphasis is on autonomous training rather than developer-facing assistance.

💡 Summary 📄 Full paper

Harness Engineering

Self-Evolving Agent Harness

Relevance: Directly addresses the harness as an object of study: it formalizes how prompts, memory, workflows, tools, and code scaffolding are revised across runs while model weights stay fixed. It also discusses validation, admission rules, and safeguards for runtime privileges, which connects to governance of harness changes. The paper is more theoretical than interactive, so human-in-the-loop observability is not its main focus.

💡 Summary 📄 Full paper

Human-in-the-Loop Evaluation

No paper recommendations for this topic.

Human-AI Collaboration

No paper recommendations for this topic.

Simulated Users

Cowork-Gym: Synthesizing Interactive Work Environments at Scale

Relevance: Includes a stateful user simulator that responds to agent actions through dialogue and direct edits to a shared workspace, aiming to capture realistic cowork dynamics. The paper uses this simulator to train and evaluate agents at scale. It does not deeply validate simulator fidelity against real users, which is a gap relative to the topic’s emphasis.

💡 Summary 📄 Full paper