Why Your ML Pipeline Needs an Agent (and How to Build One)

Name: Why Your ML Pipeline Needs an Agent (and How to Build One)
Uploaded: 2026-09-07T10:18:34.000Z
Duration: 31 min 10 s
Description: Explore the paradigm shift of Agentic ML, detailing how context-aware AI agents can autonomously manage and optimize machine learning workflows, moving beyond…

Yusuke Shibui; Prashanth Gurram

PyCon AU·Yusuke Shibui & Prashanth Gurram

31:10

Notebook

Type below to lock the video timestamp, or press S to capture a slide.

Why Your ML Pipeline Needs an Agent (and How to Build One)

Yusuke Shibui Prashanth Gurram PyCon AU

31:10·2026·graduate·Conference Talk

Explore the paradigm shift of Agentic ML, detailing how context-aware AI agents can autonomously manage and optimize machine learning workflows, moving beyond traditional AutoML approaches.

(Yusuke Shibui, Prashanth Gurram) Every ML team knows the pain: the majority of practitioner time goes into data wrangling, pipeline plumbing, and environment configuration — not solving business problems. Getting a pandas prototype into production still takes months, and that’s before you factor in monitoring, drift, and retraining. Agentic ML offers a fundamentally different approach: AI agents that understand your data context, reason about which steps to take, and execute ML workflow stages autonomously — while keeping humans in the loop for strategic decisions. In this talk, we explore why Agentic ML is a paradigm shift beyond AutoML, what makes context-aware agents effective, and how these ideas connect to Python workflows in practice. We’ll walk through real patterns for feature engineering, distributed training, and model monitoring — with honest lessons about where agents shine and where they still fall short. You’ll leave with a practical framework for thinking about agent-assisted ML in your own stack.

Systems & CS Computer Architecture Processor Design & Pipelining MLOps AI Agents Python Programming