How to Get AI to Stop Being Weird and Actually Be Useful
summary
Claims that AI does “30–50% of work” and that it will wipe out entry‑level jobs are overstated; concrete failure modes are hallucinations and agents drifting into creative actions. Biggest blockers: messy or duplicate data, using generative models for precise finite tasks, and lack of workflow adoption. Practical fixes: build an advanced prompt library, run hundreds of tracked experiments with a 1–3 success score, pick a toolchain (Azure/Gemini/OpenAI + orchestrator like LangChain), and staff forward deployment roles to co‑design agents. Goals: move one adoption level in three months.
AI-written summary · about
chapters (13)
- 0:00Welcome
- 2:10AI hype vs usefulness
- 6:41AI adoption mindsets
- 13:40Predictive vs agentic AI
- 16:05Software eras
- 18:54Why AI adoption fails
- 22:00Overcoming resistance and ROI
- 24:05AI power user
- 27:57AI-native toolkit
- 30:23Advancing AI skills
- 31:46AI adoption mistakes
- 33:28AI: I don't know
- 35:13Q&A
related talks
28mWhy AI Works for Individuals but Fails in Enterprises, and How to Fix It.
51mAssessing AI's progress
29mThe AI Teammate Is Here — Now What?
20mThe Three Ways AI Agents Fail. Here's How to Fix Yours.
27mEngineering in the Age of AI: Making AI Work for Us, Not Replace Us
41mPanel discussion: AI in code development – hype vs reality
13mAI as a Strategic Asset: Building Competitive Advantage
16mBuilding Applications with AI Agents