The Hard Truth About AI Agents in Production
summary
Demo builds a supply‑chain agent on a laptop that watches a Port of Rotterdam weather delay, analyzes 20 shipments with 14 affected, applies policy tiers (premium/platinum/silver) and surfaces a three shipments, two air slots conflict that triggers a human decision; prototype generates customer emails and runs in a day but production needs vector DB memory, orchestration, SAP integrations, observability and audit logs, and expensive on‑prem inference on H100 A100 GPUs; reliance on OpenAI/Anthropic APIs plus messy data, latency, non‑determinism and integration debt block shipping.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:23AI agent definition
- 4:32Supply chain demo
- 10:14POCs fail in production
- 11:17Enterprise adoption challenges
- 15:48AI challenges for agents
- 17:54Engineering agent systems
- 20:38Key takeaways
- 24:10Q&A
related talks
46mFullstack AI: The Building Blocks for Production Ready AI-Powered Software
40mAI Agents in Practice
11mAI took over dev. Prod is next. Here's why it has to be
31mBuilding and Orchestrating Production-ready Agentic AI Systems
30mWhy Agents Don’t Ship: Closing the Production Gap for Real Business Impact
57mWhen AI Agent Teams Work (And When They Just Create More Work)
15mThe Responsible Path to Hyperproductivity with AI Agents
45mAI agents in practice - beyond the prototype