Angie Jones is vice president of developer experience at the Agentic AI Foundation, where she advances open standards and practical engineering methods for autonomous software. An IBM Master Inventor with more than 25 patented inventions, she created Test Automation University and now applies decades of experience in software reliability, developer education, and organizational change to agentic systems.
From automation engineer to agentic infrastructure
Jones initially studied business at Tennessee State University before a programming course redirected her toward computer science. She began her engineering career at IBM, earned its Master Inventor designation, and later worked in automation roles at companies including Teradata, LexisNexis, and Twitter. Her patents span virtual worlds, collaboration software, social networking, and software-development processes.
At Applitools, she moved into developer advocacy and created Test Automation University, a free platform teaching programming and automated testing. Her public Java programming exercises and hybrid automation framework extend that educational work into inspectable, reusable code.
At Block, Jones led company-wide AI enablement across functions including engineering, marketing, design, finance, and legal. Confronting the gap between widespread coding-assistant use and actual delivery improvements, she established an AI champions program of approximately 50 engineers. Participants prepared repositories with context files, guardrails, reusable workflows, automated review, and attribution practices tailored to different engineering environments. Those foundations enabled teams to delegate work through Slack and issue trackers, run agents in isolated cloud workspaces, and coordinate changes across repositories. Her account of scaling agent adoption emphasizes organizational readiness over raw tool usage.
At the Linux Foundation-hosted Agentic AI Foundation, Jones supports open agent interoperability across technologies including the Model Context Protocol, AGENTS.md, and goose. Her mcp-selenium project connects that mission to her automation background by making Selenium WebDriver available to AI agents.
- Systems before prompts. Jones designs agents as components within larger systems of tools, workflows, data, and human oversight. Her Relocation Scout house-hunting agent separates listing normalization, neighborhood research, commute calculations, structured output, and approval decisions; explicit state tracking and idempotent retries help interrupted workflows recover safely.
- Deterministic code, model judgment, human authority. Exact calculations belong in conventional scripts, ambiguous interpretation can go to models, and consequential actions require human approval. Jones treats outside content as untrusted input and restricts agent permissions to limit potential harm.
- Structured, persistent agent memory. Her agent-memory architecture adapts Andrej Karpathy’s LLM-wiki pattern into queryable knowledge, durable action histories, and schema-governed updates that survive individual sessions.
- Agent evaluation grounded in testing discipline. Her test pyramid for AI agents combines deterministic unit tests, recorded interactions, repeated benchmarks, and rubric-based assessment to make nondeterministic behavior measurable without turning ordinary continuous integration into unstable live-model testing.
Jones also examines automation’s human consequences: engineering gains and increasingly autonomous delivery cannot be evaluated independently of their effects on workers.
