Muktesh Mishra is an applied AI engineering leader at Adobe focused on making unpredictable AI systems measurable, reliable, and continuously improvable. He coauthored CI/CD Design Patterns and is a 2026 Continuous Delivery Foundation ambassador, bringing established software-delivery practices to generative AI.
Earlier in his career, Mishra worked on automation and synthetic-data platforms at Capital One, where he also analyzed Java 10 and its accelerated release cycle. At Adobe, his work progressed from the Adobe Sensei platform to applied AI for developer platforms. His 2024 book, coauthored with Garima Bajpai, Michel Schildmeijer, and Pawel Piwosz, examines scalable delivery pipelines, deployment patterns, security, compliance, and generative AI.
Making AI evaluation operational
- Eval-driven development: Treat evaluations as continuous regression checks for changing prompts, models, and workflows, balancing automated speed against human judgment.
- Evaluation datasets as living infrastructure: Start with focused synthetic examples, label distinct application flows, and refine multiple datasets as real system behavior exposes gaps.
- Application-specific AI evaluation: Assess retrieval systems for accuracy and usefulness, generated code for functional correctness against the actual codebase, and agents for execution trajectories, multi-turn interactions, and tool-call correctness.
- Scalable evaluation pipelines: Use caching, orchestration, parallel execution, and aggregated results to make frequent measurement practical. His AI Engineer World’s Fair framework connects those techniques to customer trust and dependable product improvement.
