building reliable AI systems in production
summary
Claim: concrete systems for reliability not vague best practices. Shows a dashboard that logs every generation, prompt versioning with sandbox side-by-side diffs, and a deployed “first AI sales agent” that actually ran prospecting and outreach. Describes memory architectures (file system persistence, namespaced graphs, dreaming), tool design as product, durable workflows with pauses for approvals, and a practical trust ladder from observation to autonomy. Mentions enterprise deployment at Sycamore and a $65M seed round as evidence of scale.
AI-written summary · about
chapters (8)
- 0:00From prompts to agents
- 3:50AI startups lessons learned
- 7:41Context memory and knowledge
- 11:37Designing agent workflows
- 15:07Trust and autonomy in agents
- 19:15Orchestration and human judgment
- 22:08Evaluations and flight recorder traces
- 24:58Building reliable AI systems
related talks
24mScalable GenAI in Production
01h 06mAI in the Trenches: Engineering Real-World Systems
36mBeyond the Hype
27mHow to Build AI Agents and Keep Your Sanity
29mFrom Experiment to Enterprise: Architecting AI for Stability and Scale
54mArtificial Intelligence seen from the software development lifecycle perspective
46mFullstack AI: The Building Blocks for Production Ready AI-Powered Software
45mFrom Startup to Scale-Up: How We Build and Scale AI Product Teams at Ailylabs