Aojie (Justin) Yuan — WritingPlain-language explainers of research on trustworthy AI agents, LLM reasoning, memory, and evaluation.https://aojieyuan.com/blog/2026-10-13T00:00:00ZAojie (Justin) Yuanhttps://aojieyuan.com/Why Does Decision-Focused Learning Often Fail to Beat MSE?https://aojieyuan.com/blog/why-decision-focused-learning-often-matches-mse2026-10-08T00:00:00Z2026-10-13T00:00:00ZDecision-focused learning trains predictors on downstream decision loss, yet gains over MSE are often small. Aojie (Justin) Yuan explains Jacobian rank collapse and what the experiments show.What Makes an AI Agent Auditable?https://aojieyuan.com/blog/what-makes-an-ai-agent-auditable2026-10-08T00:00:00Z2026-10-13T00:00:00ZAccountability for AI agents requires auditability. Aojie (Justin) Yuan explains the five dimensions of agent auditability, why detect, enforce, and recover mechanisms are all needed, and what the Auditability Card is.Is a Concept Subspace Just the Logit Lens in Disguise?https://aojieyuan.com/blog/is-a-concept-subspace-just-the-logit-lens2026-10-08T00:00:00Z2026-10-13T00:00:00ZA weights-only geometric test checks whether an extracted concept subspace overlaps the model's output readout. Aojie (Justin) Yuan explains the diagnostic, its controls, and what it shows across 26 models.How Do You Retrieve Medical Images by Lesion Region, Not Just the Whole Scan?https://aojieyuan.com/blog/how-to-retrieve-medical-images-by-lesion-region2026-10-08T00:00:00Z2026-10-13T00:00:00ZAojie (Justin) Yuan explains HMAR, a mixture-of-experts medical image retrieval framework that pairs global and lesion-level experts, trains without bounding boxes, and hashes with KAN layers.Do LLMs Reason the Same Way in English, Code, and Math?https://aojieyuan.com/blog/do-llms-reason-the-same-way-in-language-code-and-math2026-10-08T00:00:00Z2026-10-13T00:00:00ZAojie (Justin) Yuan explains the Format-Agnostic Reasoning Subspace (FARS): a 10-dimensional subspace where LLMs represent the same reasoning concept across prose, code, and math — and the declarative-procedural asymmetry it reveals.Do LLMs Know When Their Chain-of-Thought Is Wrong?https://aojieyuan.com/blog/do-llms-know-when-their-reasoning-is-wrong2026-10-08T00:00:00Z2026-10-13T00:00:00ZLanguage models internally detect their own reasoning errors while sounding confident. Aojie (Justin) Yuan explains hidden error awareness in chain-of-thought — a 0.95 AUROC probe signal — and why it is diagnostic, not causal.Can Computer-Use Agents Do Visual Software Engineering?https://aojieyuan.com/blog/can-computer-use-agents-do-visual-software-engineering2026-10-08T00:00:00Z2026-10-13T00:00:00ZReal debugging means running software, looking at it, and fixing the code. Aojie (Justin) Yuan explains CUA-SWE, a benchmark where agents must combine screenshots, GUI interaction, and code edits — verified by deterministic tests.How Do You Secure Multi-User AI Agent Networks?https://aojieyuan.com/blog/how-to-secure-multi-user-ai-agent-networks2026-08-08T00:00:00Z2026-10-13T00:00:00ZWhen each person has an AI agent acting on their behalf, harm can travel between users through shared tasks. Aojie (Justin) Yuan explains cross-user agent security and how an auditable sandbox benchmark measures it — the idea behind WeClawArena.How Do AI Agents Remember Long Conversations?https://aojieyuan.com/blog/how-ai-agents-remember-long-conversations2026-08-08T00:00:00Z2026-10-13T00:00:00ZLong-term agent memory needs retrieval that is cheap when the query is easy and thorough when it is hard. Aojie (Justin) Yuan explains workload-adaptive cascade retrieval — the idea behind AgentIR.Do LLM Reasoning Models Need All Their Memory in HBM?https://aojieyuan.com/blog/do-llm-reasoning-models-need-all-memory-in-hbm2026-08-08T00:00:00Z2026-10-13T00:00:00ZLong chain-of-thought fills GPU HBM with KV cache, most of which barely matters. Aojie (Justin) Yuan explains a semantics-aware memory hierarchy that tiers the KV cache by importance — the idea behind 'Not All Thoughts Need HBM'.Why Every AI Agent Needs an Audit Layerhttps://aojieyuan.com/blog/why-ai-agents-need-an-audit-layer2026-08-07T00:00:00Z2026-10-13T00:00:00ZAn audit layer records and governs what an AI agent does — every tool call, dollar, and decision. Aojie (Justin) Yuan explains why pre-execution auditing beats post-hoc logging, and how to add one with zero code changes.What Is Charter-Governed AI Orchestration?https://aojieyuan.com/blog/what-is-charter-governed-ai-orchestration2026-08-07T00:00:00Z2026-10-13T00:00:00ZCharter-governed AI orchestration puts a machine-readable constitution — mission, budget, and rules — in charge of a multi-agent system, with a CFO that enforces spend and an auditor that scores outcomes. Aojie (Justin) Yuan explains the idea behind Sovereign-OS.What Is a Self-Evolving Verification Agent?https://aojieyuan.com/blog/what-is-a-self-evolving-verification-agent2026-08-07T00:00:00Z2026-10-13T00:00:00ZA self-evolving verification agent checks whether an AI's claims are supported by its sources, emits structured evidence and error diagnoses, and improves itself with a process reward. Aojie (Justin) Yuan explains the idea behind SEVA.Do AI Agent Skills Stay Safe Across Languages?https://aojieyuan.com/blog/do-ai-agent-skills-stay-safe-across-languages2026-08-07T00:00:00Z2026-10-13T00:00:00ZPublic AI agent skills differ by language: English skills lean technical, Chinese skills lean toward media, social, and finance apps — and safety signals are uneven. Aojie (Justin) Yuan on cross-lingual security of the agent skills ecosystem.How Do You Benchmark Tool-Use Agents for the Sim-to-Real Gap?https://aojieyuan.com/blog/benchmarking-tool-use-agents-sim-to-real-gap2026-08-07T00:00:00Z2026-10-13T00:00:00ZTool-use agents are tested on clean inputs but deployed into typos, timeouts, and duplicate tool names. Aojie (Justin) Yuan explains the sim-to-real gap for tool-use agents and how RobustBench-TC and a domain-randomized RL recipe measure and close it.