What companies are learning about AI agents in the real world
bigdata · x · 2026-11-20
Ben Lorica's Gradient Flow piece surveys agent deployments at banks, utilities, manufacturers, and retailers: building agents is getting easier, running them isn't. Key takeaways:
- Limit what agents can do, not how they reason: Duke Energy uses agents to coordinate grid-planning engineering simulations with engineers making final calls; DBS Bank chains 70-80 specialized agents to help prepare credit memos with human-owned final analysis.
- Lyft's caveat: overly scripted workflows prevent agents from handling unanticipated situations; fraud and safety issues route to humans instead of prescriptive step-by-step flows.
- Hidden prerequisite: production success often rests on pre-AI-boom investments — data cleanup, process documentation, internal software improvements.
- Most published numbers come from vendors or the companies themselves, so treat them skeptically.
More from coding & agent
- CapCut plugin connects ChatGPT to auto-cut gameplay highlights, then human picks the payoff — iamfakhrealam · 2026-11-21
- Dev wires a Runway-generated avatar into Nous Research's Hermes Agent, bashing boring agent UIs — alexcovo_eth · 2026-11-21
- Lean compiles, but the natural-language proof is still wrong — jasondeanlee · 2026-11-21
- DeepMind demos ski simulator built by Gemini 4 Neon from real Mont Blanc terrain — GoogleDeepMind · 2026-11-21
- pstack v0.15.46 trims skills, adds Gemini as panel reviewer; pinning question raised — petrusenko_max · 2026-11-21
- Hamel Husain on evals: a good process beats good tools — HamelHusain · 2026-11-21