Interrupt 2026, LangChain’s flagship agent conference, returns to The Midway in San Francisco on May 13–14, 2026. Last year’s inaugural event packed 800 builders into one venue and answered the question, “Can agents work in production?” with a definitive yes. This year’s question is different, and harder: how do you make agents work at enterprise scale?
That single shift, from “does it work” to “does it work for thousands of users, across regulated systems, with real consequences,” is the through-line of the entire 2026 agenda. We’ve put this guide together to help AI engineers and AI executives spend their two days at The Midway well.
What’s different about Interrupt 2026
The 2026 agenda is dominated by production case studies from companies running agents at scale: Toyota, LATAM Airlines, Honeywell, Coinbase, Lyft, LinkedIn, Rippling, monday.com, Etsy, Bridgewater Associates, Clay, and Abridge. The themes that recur across sessions are evals, multi-agent architectures, and the production-scale tradeoffs of moving past the demo phase.
If you’re building agents inside an enterprise, or selling to one, those production talks are the highest-signal sessions on the schedule.
Must-see sessions for AI engineers
These are the sessions where you’ll learn the most about actually shipping reliable agents.
Day 1
Evaluating Agents at Scale. Nick Ung, Lyft (Day 1, 12:00 PM). Lyft’s Safety and Customer Care team is one of the few groups talking publicly about closing the feedback loop between failed traces, ops, and engineering.
How We Built ToyotaGPT. Kordel France and Ravi Chandu Ummadisetti, Toyota (Day 1, 11:40 AM). A look at how a large industrial company architected an internal agent platform.
Compliance That Ships: Building Jade. Philipp Comans (Day 1, 1:40 PM). Compliance is where most enterprise agent projects stall. This talk addresses it head on.
Day 2
How Coinbase Scaled AI Support with a Multi-Agent System. Evan Kormos (Day 2, 3:10 PM). Multi-agent systems still trip up most teams. Coinbase has one running at financial-services scale.
Agent Evaluations. Shreya Shankar and Hamel Husain (Day 2, 3:30 PM). Two of the most rigorous voices in agent evaluation, on the same stage. Expect a methodology-heavy session.
Must-see sessions for AI executives
If you’re setting AI strategy or buying enterprise AI, prioritize these.
Future of AI Agents. Andrew Ng with Harrison Chase (Day 2, 1:00 PM). Ng’s read on where agents are headed tends to shape the next quarter of board conversations.
Agents in the Enterprise. Aaron Levie (Box CEO) with Harrison Chase (Day 1, 4:10 PM). Box has been ahead of most enterprise software companies in committing to AI-native workflows.
Building Frontier CX Agents. Carlos Pereira, Cisco (Day 1, 10:30 AM). Cisco is the presenting sponsor and has scaled customer experience agents to a base most companies can only dream of.
Agents in the Enterprise. CJ Desai (MongoDB CEO) with Harrison Chase (Day 2, 1:30 PM). What the data layer looks like when agents move from experiments to production systems.
Agent infra questions to explore @ Interrupt
We’ve watched enough enterprise teams vet agent infrastructure to suggest a short framework for the action layer, the part of the stack where reasoning becomes real work. Whether you’re walking the sponsor floor or grading a session, ask:
- Is authorization enforced at runtime? Can the agent prove that this user is allowed to perform this action on this system at the moment of the call? OAuth flows, token lifecycle, and least-privilege scoping at the tool level are where most agents get blocked by security.
- Is tool execution reliable? “Update the Acme deal” needs to resolve to the right opportunity ID. Tools should be agent-optimized for LLM intent, with retries, validation, and testing built in, so the agent never has to reason about schemas, IDs, or pagination.
- Can you govern every action? Audit trail, OpenTelemetry-compatible observability, version control for safe upgrades, project- and org-level segmentation. If a vendor can’t show you these, they’re still in prototype territory.
- Do deployment options comply with my requirements? Cloud is fine for some workloads. Regulated verticals like finance, healthcare, and government usually need self-hosted or air-gapped. Ask about Helm and Docker.
- Will this lock me in? Any LLM, any framework, any agent client. The AI era requires agility. If you can’t swap models without rewriting tools, you’re locked in.
A note on why Arcade.dev is at Interrupt
Arcade.dev is at Interrupt because the conference is converging on the action layer, the same problem we built our MCP runtime to solve. Stop by our booth if you want to talk through delegated OAuth for agents, the 7,500+ tools in our catalog, or how to think about an agent control plane. Or book a meeting with the team to see Arcade in action.
FAQ
When is LangChain Interrupt 2026? May 13–14, 2026.
Where is Interrupt 2026 held? The Midway, 900 Marin St., in San Francisco’s Dogpatch neighborhood.
Who is Interrupt 2026 for? AI engineers, AI researchers, and engineering and product leaders building or buying agentic AI applications at enterprise scale.
Who are the keynote speakers? Harrison Chase (LangChain), Andrew Ng (DeepLearning.AI), CJ Desai (MongoDB), Aaron Levie (Box), and Carlos Pereira (Cisco), plus production talks from teams at Toyota, Lyft, LinkedIn, Coinbase, Honeywell, Rippling, monday.com, LATAM Airlines, Etsy, Clay, Abridge, and Bridgewater Associates.
Who is the presenting sponsor? Cisco Customer Experience. Other sponsors include Arcade, Box, CopilotKit, DE Shaw, Elastic, Factory, Fireworks, Focused, Kong, Merge, MongoDB, Oracle, Parallel, Redis, Tavily, and Vapi.
