The AI in Business Podcast Why Enterprise AI Fails Without a Context Engine - with Eran Yahav of Tabnine
19 snips
Mar 25, 2026 Eran Yahav, CTO and co-founder at Tabnine, builds enterprise context engines to give AI systems organizational memory and reasoning. He discusses why AI struggles in legacy environments, a three-layer architecture (LLM, agent UI, context engine), and how mapping dependencies, precomputation, and running inside security perimeters can boost reliability and cut costs.
AI Snips
Chapters
Transcript
Episode notes
Organizational Understanding Explains 80% Agent Failure
- AI agents fail ~80% on complex enterprise tasks because they lack organizational understanding.
- Eran Yahav compares agents to new hires who need months of onboarding to navigate brownfield legacy systems with undocumented business logic.
Three Part Agent Stack Needs A Context Engine
- Effective enterprise agent stacks require three parts: the LLM, the task agent/interface, and a context engine.
- The context engine connects to internal systems and supplies the enterprise-specific information the LLM lacks.
Prebuild Dependency Maps For Agents
- Pre-compute and maintain a dependency map so task agents don't rediscover relationships for every task.
- Eran describes context agents that crawl systems, build dependency maps, and expose focused surfaces for task agents to inspect.