Live session - Why Good AI Still Gives Incomplete Answers and How to Fix It
Even the most capable AI models can give incomplete or confidently wrong answers. Why does that happen? Usually the AI can't reach the right information, or it's built to make you happy instead of being right. In this session, we cover: 1. The Live Smart Agent: Watch autonomous multistep retrieval plan, decompose, evaluate and iterate across Knowledge Boxes, MCP-connected apps and web sources in real time. Learn more - https://www.progress.com/agentic-rag/features/smart-agent 2. Teams, without the rebuild: The same indexed content, access controls and source-cited answers your team already built, now reachable inside Microsoft Teams as a configuration change, not a front-end project. 3. WordPress ingestion that keeps what crawling throws away: See the difference between crawl-based discovery and source-side ingestion—structure, taxonomies and metadata intact. 4. What governed autonomy actually looks like: Source citations, REMi quality scoring and audit logs attached to multistep behavior from the start, not bolted on after. 5. Why extensibility is the real feature this quarter: How one existing deployment inherits all three capabilities with no migration, and what that means for your own roadmap. Learn more - https://www.progress.com/agentic-rag Chapters 0:00 Welcome & intros 2:34 Why capable AI still gives wrong answers 3:06 What is AI sycophancy? 5:08 Retrieval problems search indexes can't fix 6:11 Why you can't just send 1M documents to ChatGPT 8:07 What breaks DIY RAG systems 10:02 Where simple RAG falls short 10:39 How RAG works: indexing & retrieval 13:13 What is agentic retrieval? 14:46 How queries get broken into sub-queries 17:55 Context relevance & re-querying for gaps 19:14 Setting up agentic retrieval with MCP 20:50 Use cases: sales, support, compliance, engineering 22:35 Microsoft Teams app delivery 23:43 WordPress plugin for easier ingestion 26:00 Live demo: ingesting data & live sync 27:15 Connecting an MCP server 28:24 Testing search with citations 29:24 Reactive vs. Plan & Execute mode 30:36 Picking models: Rephrase, Executor, Summarize 33:09 50+ LLMs & side-by-side testing 34:09 Custom app demo: ERP + OpenEdge 37:43 #1 on the Snowflake MADQA leaderboard 38:39 Customer story: Scrivens 40:39 Q&A: agentic retrieval vs. simple RAG performance 43:16 Q&A: when to use simple RAG vs. agentic 43:57 Q&A: Starter vs. Pro plans 45:50 What's next: decision models & custom indexing