Elvin Aghammadzada is a junior data science engineer at DataRobot developing agent skills that teach coding agents to operate enterprise AI platforms. His work treats platform expertise as software: modular instructions that agents can discover, apply, evaluate, and maintain.
Aghammadzada studied at Northeastern University and developed projects involving machine learning and immersive virtual tours. His early writing examined hindsight bias in language models, showing how knowledge of an outcome can distort assessments of whether a decision was sound. He also argued that burdensome AI regulation can disadvantage smaller companies. In 2025, he helped design an AI-focused summer program for Boston youth and led workshops.
At DataRobot, he coauthored a practical guide to platform skills for Claude Code. The company’s datarobot-agent-skills repository extends this approach to model training, deployment, prediction, monitoring, and data preparation across multiple coding-agent environments.
- Context engineering: Agents perform worse when overloaded with documentation, tool definitions, accumulated errors, and irrelevant results. Aghammadzada treats context as a finite engineering budget.
- Progressive disclosure: Skills expose concise metadata first, load detailed instructions only when needed, and can execute supporting scripts without inserting their complete implementations into context.
- Platform teachability: Enterprise products gain a fluency moat when agents can quickly learn their workflows and convert user intent into reliable outcomes. Governance, authentication, tracing, and integration remain essential.
- Complementary infrastructure: Skills provide procedural knowledge, while the Model Context Protocol can supply remote execution, protected-system access, authentication, and additional computing resources.
- Production-grade safeguards: Reusable skills need versioning, testing, and evaluation, alongside protections against prompt injection, unverified marketplaces, and unsafe execution.
His AI Engineer World’s Fair 2026 talk frames these concerns as a fundamental shift: enterprise platforms must become understandable to agents, not merely accessible through APIs.
