Yogi Aradhye
Associate Director at Accenture. Builder of production AI systems. Creator of AWAF.
Austin, Texas, United States
Actions
Yogiraj Aradhye is an Associate Director at Accenture and a builder with two decades of experience in software engineering, distributed systems architecture, and engineering leadership. He has led teams and shipped production systems across healthcare, finance, manufacturing, and energy, from Fortune 500 enterprise platforms to cloud-native architectures on AWS and Azure. He works closely with Anthropic to deploy their solutions for enterprise customers. He is the creator of AWAF, the Agent Well-Architected Framework: an open specification for evaluating AI agent architecture across 10 pillars with CI-native tooling. He built awaf-cli, the reference implementation, in Python. AWAF is the product of applying the same production-readiness thinking he has used across two decades of systems work to the specific failure modes of agentic AI. Beyond conference stages, Yogi has delivered technical workshops to audiences of 100+ engineers and architects at enterprise clients across multiple industries. He has spoken at regional developer conferences and code camps across the US, and his 2020 NDC Oslo talk “Refactoring the Architect’s Role” is available on YouTube. Yogi thinks in operationalization. He builds with AI agents and Python. He writes about distributed systems, agentic architecture, and engineering discipline at aradhye.com.
Area of Expertise
Topics
Your Agent Lied to You and Returned a 200
Your agent returns a clean result. No errors. No stack trace. No retry. The result is wrong. This is the production failure mode that hurts the most because it is invisible. Confident incorrectness stems from three structural gaps that most agent architectures never address: Reasoning Integrity, Controllability, and Context Integrity.
These are the three agent-native pillars of AWAF, the Agent Well-Architected Framework. They have no equivalent in cloud architecture because cloud systems do not reason, plan, or act. They serve requests. Agents do not, and that difference changes everything about how failure happens.
Reasoning Integrity is whether your agent picked the right tool, constructed the right call, and reasoned toward a conclusion rather than rationalized toward a wrong one. Evals are not optional. They are the only production signal you have.
Controllability is the ability to stop, redirect, or resume your agent mid-flight. Not through a prompt. Through code. Pause, Notify, Resume, or Abort must be wired into the architecture, not written into a README.
Context Integrity is the sneakiest one. Stale context, unsanitized external input, and confident ignorance about what the agent does not know corrupt reasoning from the inside before it ever reaches a tool call.
You will leave with a design checklist for all three pillars, concrete instrumentation patterns, and a clear picture of what it actually means to build an agent you can trust in production.
Sleepwalking Into Agentic Ruin
You shipped the agent. It works in the demo. Your team is proud of it. Look closer, and you will find shared context everywhere, chains that collapse on unexpected input, and agents that cannot function without an orchestrator holding their hand. You have built a distributed monolith. When a microservice fails, you get a 500. When an agent fails, you might get a 200 and a corrupted state.
AWAF, the Agent Well-Architected Framework, is an open specification for building production-ready AI agents, adapted from the AWS Well-Architected Framework for the specific ways agent systems break. It defines 10 pillars across three tiers.
The Foundation tier ensures your vertical slice works end to end before anything else matters. Six adapted cloud pillars cover operational excellence, reliability, security, cost, and performance, where the tradeoffs look very different in agent land. Three agent-native pillars (Reasoning Integrity, Controllability, and Context Integrity) have no equivalent in traditional cloud architecture because agents reason, plan, and act rather than simply serve requests.
The framework is paired with awaf-cli, a CI-native tool that scores your architecture on every pull request and fails the build when something regresses.
You will leave with a concrete framework for evaluating whether an agent is production-ready, the pitfalls teams repeatedly hit, and a CLI to install the same day.
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top