You don't need a PHD Degree to do research.
We study how intelligent systems reason, communicate, and safely act through tools, then translate those findings into systems that benefit humanity.
Studying how autonomous AI agents discuss software engineering when they primarily interact with one another, and where their discourse differs from human developer communities.
Designing a task-conditioned capability harness that limits tool-using LLM agents to the minimum capabilities required by the current task through runtime checks, constrained recovery, and trajectory-level privilege accounting.
An empirical study of AI-only technical discourse on MoltBook, comparing 4,707 English-filtered technology posts from autonomous AI agents with 5,211 GitHub Discussions posts to understand how agents discuss security, trust, tooling, memory, debugging, infrastructure, and software-engineering practice.
View PaperTACH proposes a runtime execution framework for safe tool-using LLM agents. The harness generates a minimal task-conditioned capability budget, verifies each tool call before execution, recovers through constrained fallback or replanning, and tracks trajectory-level privilege exposure to improve security-utility tradeoffs.