Banjo Obayomi is a software engineer, solutions architect, and developer advocate who builds AI systems that operate browsers, interpret cloud infrastructure, and turn natural-language instructions into practical software workflows. His work spans reproducible data engineering, cybersecurity research, reinforcement learning, and the emerging architecture of autonomous web agents.
Obayomi earned computer science degrees from the University of Maryland, College Park, in 2011 and Loyola University Maryland in 2015. At Two Six Labs, where he became a senior research engineer, he developed serverless machine-learning applications, experimented with reinforcement learning for Unity’s Obstacle Tower Challenge, and built an automated AWS knowledge graph from technical documentation.
He also coauthored Magicwand, a research framework for generating reproducible, labeled datasets of low-and-slow distributed denial-of-service attacks. His O’Reilly contribution to 97 Things Every Cloud Engineer Should Know advocated reproducible data infrastructure, applying software-engineering practices such as automation and versioning to data pipelines. He also cohosted the DataOps Podcast.
By 2021, Obayomi was a senior developer advocate at Amazon Web Services. His subsequent work on AI-assisted solutions architecture explored systems capable of answering cloud questions, retrieving documentation, and generating infrastructure diagrams. His independently developed ChatAWS applied conversational AI to creating websites and AWS Lambda functions.
How he builds agents that take action
- Natural-language browser automation: Using Amazon Nova Act, Obayomi designs agents that identify visible interface elements and navigate websites from written instructions. Demonstrations have compared products across parallel browser sessions and located highly rated coffee shops near a particular hotel.
- Portable browser tools through MCP: He created a Model Context Protocol server exposing browser sessions, actions, screenshots, and parallel tasks as reusable tools. An external language model handles planning while Nova Act executes browser interactions; the same server can connect to different AI clients.
- Multi-agent cloud architecture: With Strands Agents, he combines specialized architecture, pricing, and presentation agents to assemble migration recommendations, current cost estimates, diagrams, and executive-ready slides. His approach favors concise, code-first orchestration and reusable tools.
- Practical limits on autonomy: Obayomi openly demonstrates failures involving CAPTCHAs, ambiguous instructions, stalled navigation, and unexpected browser state. He emphasizes detailed tool descriptions, execution logs, human intervention when needed, and workflows that respect security controls.
At AI Engineer World’s Fair 2025, he and Du’An Lightfoot brought these ideas together in a hands-on workshop on Nova Act, MCP, and Strands Agents, showing how agent systems can move from a written request to observable action while exposing their operational limitations.
