이 에피소드에 관해
On Technically Sound Strategy, host Chip LaFleur sits down with Bob Flavin, EVP at LaFleur, to get practical about AI agents. Everyone is talking about them. Few people explain how they get used day to day. Their rule at LaFleur: robots do robot work, and people do people work.
Flavin explains the gap between using AI as a chatbot and running an agentic workflow. With a chatbot, the human drives every step. With an agent, you train it once on purpose and context, give it a trigger like a weekly schedule, and let it return finished work with its reasoning. His example is SEO and GEO work. An agent connected to GA4 and Google Search Console can surface content opportunities, rewrites, and decaying pages, which frees the team for the work only humans do well.
The two dig into what makes agents reliable. That means a clear definition of what good looks like, evaluation loops that help the agent learn from its misses, and regular audits of agent memory so early mistakes don't carry into later work. They also cover guardrails. Flavin doesn't let his agents send or publish anything without his approval, especially in regulated industries where you can't put the genie back in the bottle. They also talk through programmatic checks, secondary review agents, and a "panel of experts" trained on state bar rules that scores a draft before a person ever sees it.
They close with where this is headed. Capabilities will keep racing ahead, but adoption will likely move slower than Silicon Valley expects. A one-person, million-dollar company run by agents is no longer a stretch. The takeaway: build agents for a real business purpose, set the guardrails early, and keep a human at the point where judgment matters.
Chapters
00:00 Robot Work vs. People Work 01:00 Using AI vs. Agentic Workflows 03:00 Claude vs. an Agent 04:00 Purpose, Context, and Triggers 06:00 Defining "What Good Looks Like" 07:00 Evaluation Loops That Learn 08:00 Choosing an Agent Harness 09:00 Auditing Agent Memory 10:00 Data Hygiene and Early Draft Errors 11:00 The Memory Black Box 12:00 Guardrails on Access and Actions 13:00 Why a Human Approves Before Publishing 14:00 Live Chatbots and Programmatic Checks 15:00 Grading Agent Output 16:00 How Skeptical Should We Be of Bots 17:00 A Panel of Experts for Legal Marketing 18:00 Scoring Drafts to 90 19:00 Where Agents Go Next 20:00 Capability vs. Adoption 21:00 The One-Person Million-Dollar Company 22:00 AI Slop and Agents Policing Agents 23:00 Wrap Up
