Enterprise AI in Production

This event is not a traditional conference.

We are inverting the industry standard of 80% slides and 20% substance, prioritizing executable knowledge and real-world architectures. Submissions must adhere to our core principle: If you can show it running in production, show it.

The unifying theme across all sessions is “How do you know it’s working?” Submissions must answer this question with concrete details drawn from:

• Evaluation pipelines and methods
• Observability and tracing
• Cost data and benchmarks
• Real architecture and engineering decisions

We are seeking submissions for three primary formats:

1. Anchor talks - deep talks of 25 minutes duration + 5 minutes Q&A
   Anchor Talks set the intellectual frame for the day by addressing the “production gap”. We are looking for CTOs or principal engineers with a live system to share the real architecture of a production system.

• Workflow Orchestration: Moving from standalone chatbots to deep integration with legacy ERP/CRM systems via event-driven architectures.
• Sovereign & Localized AI: Architectures for air-gapped, regulated, or high-security environments where cloud APIs are not an option.
  Focus: What is working, what was thrown away, and the one decision you would make differently.
  Requirement: No slides featuring roadmaps.

2. Live demos - 15 minutes each
   These sessions are the core of the conference (30% of content) and must be genuine, live demonstrations of production-ready AI features. We are looking for submissions that tackle hard technical problems, including:

• Agentic Systems: multi-agent workflows with visible execution traces, showing failure and escalation live.
• Evaluation in CI/CD: demonstrating an evaluation pipeline catching a real regression (e.g., a “harmless” prompt change breaking a faithfulness eval and blocking a CI merge).
• Retrieval-Augmented Generation (RAG): live query across an enterprise corpus, comparing hybrid retrieval vs. naive RAG, and showing hallucination reduction as a metric.
• Multimodal pipelines: unifying invoices, contracts, and images in one retrieval system with sub-second query speed and live source attribution.
• AI for development: demonstrating AI code review and auto-PR workflows, where an agent generates tests, documents changes, and creates a PR with a plain-English diff explanation.
• Judge models: a live reveal of LLM-as-a-Judge bias (e.g., score flipping when answers are swapped) and the applied mitigation.
• Agentic Protocols: Live implementation of communication standards to manage multi-agent handshakes and stateful memory.
• Defensive AI in CI/CD: Automated red teaming and policy-as-code filters that block non-compliant outputs in real-time.
• Reliable Agent Skills: A live demo of “Tool Use” or “Function Calling” at scale—specifically how agents discover, select, and recover from failures across hundreds of internal APIs.

3. Lightning talks - 12 minutes each
Lightning talks are short, sharp takes, structured around one orientation slide, then the thing itself. We are seeking practitioners who can deliver maximum insight in minimum time.
Focus areas:

1. Architecture & governance: running a global AI program (e.g., GCC perspective on governance handshakes with HQ).
2. Generalizable patterns: RAG or other patterns that generalize across enterprise client engagements (GSI perspective), and the ones that never do.
3. Cost reality: real inference cost data from a production workload (e.g., GPT-4o vs fine-tuned SLM), where the audience votes before the reveal.
4. Outcome Economics: Moving beyond cost-per-token to measuring the “unit economics” of an AI action and true business ROI.
5. Skill Composition: Patterns for building “composable” agent skills that can be reused across different departments without re-training or hard-coding logic.
6. Governing the ADLC: How do we handle “Pull Requests” and “Unit Tests” when the output is generated by an autonomous agent

Startup Showcase: A live product demo presented to a room of practitioners - strictly no pitch deck.

Birds of Feather (BOF) sessions

1. If you have a hard unsolved problem, bring it to the BoF. While the final parallel BoF rooms are set by attendee voting on the day, we invite submissions for high-value discussion topics that could cluster organically, such as:
2. Agent systems that broke in unexpected ways
3. Evals that are effective in production
4. Making the CFO understand GPU costs
5. The engineering specifics of good AI governance
6. RAG vs. fine-tuning—the conditions that change the answer
7. Managing Agent Sprawl: Preventing cascading logic loops in complex multi-agent systems.
8. AI Supply Chain Security: Verifying the integrity of model weights and data in an open-source/fine-tuned ecosystem.
9. Agent Skill Discovery: The challenge of “Skill Sprawl”—how do agents know what tools are available in a massive enterprise ecosystem