🛠️ Why Voice AI Agents Fail: A Field Guide

On May 26th, we are gathering Chicago’s most active engineers, founders, and operators at Drive Capital to discuss the engineering reality of voice AI. While voice agents in 2026 often demo beautifully, they frequently fail in production due to complex turn-taking, state management, and the challenge of separating human signal from noise.

Most voice agent failures are not model failures; they are failures in the orchestration layer. This session features Rod Rivera from Rasa, who will walk through four specific production failures, demonstrate them with running code, and share an open-source repository to help you reproduce and solve these challenges in your own stack.

🗣️ Featured Speaker

Rod Rivera, Developer Relations at Rasa. With 15+ years in applied ML, Rod focuses on the practical engineering required to move beyond “audio-glued-to-text” and into resilient, production-grade voice interfaces.

🎙️ Show Your Work: Call for Demos

We are looking for active builders to present technical deep-dives or demos. We prioritize code over slides and engineering over sales pitches. If you have solved complex state management challenges or built novel agentic workflows, we want to see it.

Submit Your Demo Proposal Here

🗓️ Schedule and Logistics

• Date: Tuesday, May 26, 2026
• Time: 5:30 PM - 8:30 PM
• Location: Drive Capital, Chicago (Exact address provided upon approval)
• Format: Technical deep-dive followed by networking and community demos.

🛡️ Curation Policy

Attendance is strictly limited to 100 practitioners. We screen every registration to ensure the room is filled with engineers and researchers who are actively shipping. Space is limited and our Chicago events consistently run with a waitlist.

🤝 Our Partners

Drive Capital is a venture capital firm managing $2.2B in assets, partnering with founders across America to build market-defining companies from seed through IPO.

Rasa provides the infrastructure for enterprises to build and deploy sophisticated, mission-critical AI agents that users can trust.

Event photos

AI Tinkerers - Chicago

AI Tinkerers - Chicago

AI Tinkerers - Chicago

AI Tinkerers - Chicago

AI Tinkerers - Chicago

AI Tinkerers - Chicago

📊 AI Tinkerers Chicago Stats

• Attendees: This exclusive community of 2,198 AI professionals features a highly technical membership, with 80% specializing in Python-driven machine learning, 65% in generative AI and LLM orchestration, and 40% in cloud infrastructure. Notably, over 30% of members are active startup founders or CTOs, bridging academic excellence from UChicago and Northwestern with enterprise execution at firms like Google and Stripe.
• Companies Represented: Featuring tech giants like Google, Microsoft, NVIDIA, and Apple, alongside prominent platforms like DoorDash, TikTok, Waymo, and Robinhood, and innovative startups such as ElevenLabs, Pinecone, Hume AI, and HumanLayer, and more
• Demos: 184 demos have been submitted and 107 have been presented, with especially engaging coverage of agentic and multi-agent architectures, reliable structured generation/tool calling, and multiple flavors of retrieval-augmented generation (including graph and dynamic filtering approaches). Attendees also explored workflow automation pipelines, evaluation/observability for agents, and performance/cost optimization for real-world deployments, plus multimodal generation and structured data extraction.
• Testimonials:

  “Please continue organizing interactive AI events with industry professionals and real-world discussions about technology trends, career opportunities, and practical AI applications. I especially appreciated the opportunity to ask questions, network with others, and learn directly from experienced speakers. Events like this are very helpful for students and professionals looking to grow their technical skills and better understand the future of AI and technology.”

  — QA engineer, cloud support on May 20, 2026

  “I enjoyed the breakout room like discussions from the previous Meetup hosted by Zapier. Those allowed us to easily connect and talk about experiences.”

  — Applied AI engineer on June 10, 2026

  “The breakout groups were great.”

  — Applied AI engineer on May 20, 2026

  “Keep up the good work. This group is a really great way to stay connected to what's happening with AI in Chicago.”

  — Founder, crypto AI on April 15, 2026

  “Keep hosting more events!”

  — Data scientist @ healthcare on November 02, 2025

  “Continue keeping things technical”

  — AI engineer, fintech on March 27, 2026

  “Please keep organizing highly technical, builder-focused AI events like this in Chicago.”

  — R&D engineer, defense tech on May 27, 2026

A great demo shows working software or verifiable system behavior, not just a promising concept. Follow a builder-first style: clearly describe the specific technical challenge you solved, then demonstrate the mechanism in action (e.g., how the system behaves on different real cases, how it handles missing information, how components “contract” with each other). Audience members respond strongly to demos that produce inspectable artifacts—logs/graphs, structured intermediate state, or editor-ready project models—because it lets others understand what to copy and where trade-offs live. They also like demos that translate complexity into an approachable but still technical framing (as in a strong analogy) while maintaining architectural substance. Avoid slide-deck/marketing-pitch vibes: if you can’t show code/workflow details, failure-mode handling, or a tangible output artifact, ratings tend to suffer. Also ensure delivery quality (audibility/clarity), since even an interesting technical demo can underperform if the audience can’t follow it.

In Chicago, Luis Cisneros’ Teaching a Clinical Multi-Agent System to Say I Don’t Know earned a perfect 5/5 from two raters because it demonstrates a clinically grounded multi-agent system that abstains rather than guessing: the demo walks through multiple real cases, showing how agents converge, how disagreement is surfaced to the clinician, and how out-of-distribution cases are formally routed to human review—plus promised access to orchestration graphs, conflict logs, and the CCR tolerance-band math. Also in Chicago, Katarina Coates’ Training AI Like a Dog: What Behavioral Science Reveals About Model Alignment landed well (4/5) with audience feedback calling out the “dog-training analogy” as relatable, while still describing a real architectural idea: a behavioral auditing/intervention layer built alongside a local small model that targets root causes of drift rather than surface suppression. Finally, Pat Narendra’s Building an Agent-Human Interactive Video Studio received 4/5 (single rater) because it emphasizes a durable, inspectable human-in-the-loop workflow: agents produce scene-first structured files (script/visual intent/assets/narration settings/approval state), humans approve per-scene changes, and a deterministic renderer only consumes approved data—positioning video as “code” with resumable, debuggable, repairable outputs.