Why Your Agents Fail. Office Hours.

You can build an agent that demos beautifully. Production is a different story.
Most teams in 2026 are tuning prompts and swapping models. They're spending their time on the wrong layer. The hard part isn't the model. The hard part is the engineering around it: the state machines, the validators, the manifests, the gates. The harness.

This is a weekly Friday office hours session where Rod Rivera answers your questions about why agents fail in production and how to fix them. Submit a question when you register. Rod selects questions and answers them live. The session is recorded and published on Rasa and Prof Rod's YouTube channels afterwards.

The conversation usually covers harness engineering, state machines, validators, gates, retry logic, observability, plus the failure patterns that reappear across voice, chat, and document agents, the cost and latency engineering that demo videos always skip, and the specific reasons agent demos fall apart in production.

Questions from Nebius Academy students, builders from the broader community, and senior practitioners are all welcome.

This is the live companion to Why Agents Fail, the free course launching this summer at Rasa University and profrod.ai.

The Saturday May 2 CDTM masterclass was Episode 0. The weekly office hours are how the course conversation continues. They are a place to bring the failures you're actually hitting and get them discussed in public.

Format Sixty minutes. Friday 16:00 UK time. Live on Zoom, recorded. The first 50 minutes are questions and answers. The closing 10 minutes are Rod's observations on a pattern he's seeing across the field. Questions submitted in advance get priority. Live attendees can also unmute and ask follow-ups. You don't need to attend live to submit a question. You don't need to submit a question to attend. Recordings go up on the Prof Rod YouTube channel within 48 hours.

What you'll walk out with each week

• A working answer to the specific failure mode you're hitting, with a pattern to apply
• The vocabulary for naming what's actually wrong, "harness failures" vs "model failures," and why the distinction changes how you debug
• One tactical thing to try on Monday morning
• Awareness of what other practitioners are running into, since the questions surface what's going wrong across many different agent shapes at once

Who this is for
Builders working with AI agents in production or near-production. Nebius Academy students working through the curriculum. Developers exploring local-first AI systems. Senior practitioners running into the gap between what works in demos and what survives in production. People who want to learn by listening to questions that aren't theirs.
No preparation required. Bring any questions.

Hosted by Rod Rivera
Rod Rivera is Developer Relations Engineer at Rasa, Lecturer at Nebius Academy, Professor of the Practice at ITAM, and the educator behind Prof Rod and Sovereign Agents. He has spent 15+ years building production ML systems across Alibaba Cloud, Huawei, Samsung, Rocket Internet, and Rasa.

This series is made possible by Rasa, Nebius Academy, and Prof Rod.