S11E07 – Is AI a Bubble? Agentic AI and Privacy

Episode Summary

Mike began with a story about his Tesla’s 12-volt battery dying on the way to the studio, then spent the show on AI and money. He talked about how people and small businesses are saving money with AI, compared today’s AI boom with the dot-com crash, and explained what “agentic” AI is and how it differs from a chatbot.

The last stretch was a privacy warning: the risks of DeepSeek, of putting sensitive data into any cloud AI, and of agent tools such as OpenClaw that run on your own computer with your access.

Topics Covered

AI that saves time and money

Mike said ChatGPT’s newest version is being described as an AI worker rather than an assistant. He gave meal planning as an example: take a photo of the fridge, give it your recipes, and get a shopping list sorted by store area in minutes. He said studies show this can save people 20 to 30 percent on groceries, and that small businesses can replace $1,500 to $5,000 a month of advertising and social media management with under $150 in AI tooling. A client of his rebuilt a roughly $400-a-month scheduling product on her own with ChatGPT in a couple of hours.

Is AI a bubble?

Mike compared AI to the dot-com crash, noting that Amazon’s stock fell about 95 percent and later rose enormously, while Pets.com went bankrupt nine months after its IPO. He said some AI companies will fail, that OpenAI has never been profitable, and that legal cases over chatbot advice are coming, but that AI as a whole will survive just as the internet did. He said AI reached the same share of users in about two months, versus 10 years for the PC and 7 for the internet.

Anthropic and agentic AI

Mike said he thinks Anthropic is best placed to survive and that its higher-priced plans are much better for development and research. He explained that agentic AI means you give it a task and it carries it out, instead of just answering questions. His examples were planning a Tucson to Sedona trip to a budget and schedule, drafting a refund complaint reply, finding a Medicare doctor accepting patients, and finding subscription creep in your spending. He said the skill that matters is clearly describing what you want.

Smart thermostat savings

Mike said switching to a smart thermostat gave him about a 30 percent electricity saving. In his new home with solar, he said summer bills dropped from about $300 a month to a bit under $100, though he noted the summer had been mild so far.

DeepSeek, privacy and OpenClaw

Mike said DeepSeek is far cheaper per token than US models and did acceptably in his tests, but advised against giving it sensitive data because it is a China-based service. He said no cloud AI is truly suited to secrets, and that his office uses a local Ollama model with no internet access to strip passwords and sensitive data before anything goes to internet-connected AI. He warned that OpenClaw runs on your computer with your access and should be sandboxed. He told of a researcher whose agent, after his blood sugar dropped and he did not answer, emailed his wife that he was probably dead. He compared such tools to teenagers behind the wheel.