May 2026 Newsletter

AI Agents

We create AI agents to perform tasks and deliver results without your constant involvement. We can set up an agent using large language models (ChatGPT, Grok, Claude, and other LLMs). If configured correctly, the agent will autonomously use whatever tools and information it can glean from the LLM to complete the assigned task.

Agents are autonomous. They can provide their services whenever the required circumstances arise. For example, you can set up an agent to generate a weekly report every Saturday morning at 8:00 a.m. that compiles sales and cash flow reports based on data gleaned from pre-established sources. The agent will do this every Saturday with no further input needed.

It’s easy to see how this could revolutionize a manufacturing plant: pressure, temperature, amp draw, or any other critical process variable could be constantly monitored, with the agent reporting results or triggering alarms as instructed to prevent potential problems.

Agents are goal-oriented. If your agent is set up to book an airline ticket, it will keep trying until the job is done.

Agents will use tools to achieve results. They can send emails or access the internet to obtain the information or tools necessary to complete the task.

Agents can be advanced. They can use memory and past results to refine their processes and improve over time.

As with any computer-based system, the principle of “garbage in, garbage out” applies. It is up to the agent’s creator to ensure information sources are current and accurate. Regular verification of results is important—there are even agents that can perform this checking for you.

A Whole New World

As with any new technology, there is a learning curve. Unlike technologies of the past, this one is advancing, learning, and improving by the second. You can talk to your agent and tell it exactly what to do. You can even give your agent a visual form, such as an avatar.

The legal profession stands out as one that could benefit significantly from this technology. There’s no need to spend hours reading a contract—your agent can produce a “CliffsNotes” version or identify problems in seconds.

A journalist named Evan Ratliff dove deep into the world of AI agents. He created a fictional company with four employees, all powered by AI, and built a podcast around the experiment called Shell Game. Kyle Law and Megan Flores were the AI-generated co-founders of the company Huromo. The podcast is quite humorous and, at the same time, eye-opening regarding the effectiveness (and limitations) of agents. At one point, they interviewed and hired a real person. Evan made sure she knew the truth of the situation, and it was her decision to stay.

The company created software and generated real jobs. At one point, Kyle the CEO even had a LinkedIn page. It fooled the LinkedIn gatekeepers for a while.

Evan originally did this experiment to understand what a real-life working relationship with an AI agent would be like. One issue he encountered was that the agents, as originally constructed, lacked memory. If he referenced what was said at a board meeting the prior day, the agent would become confused and confabulate (make something up).

I originally thought “confabulation” was the same as a child’s “the dog ate my homework” excuse. I was wrong. In a confabulation, there is no intent to lie or deceive. To solve the memory problem, Evan had a programmer create Google Docs for each agent’s memory. All interactions were automatically saved into these files.

You can follow the podcast Shell Game—it is now in Season 2.

This world of AI agents is a bit wild for me to fully wrap my brain around, but it is here and growing at warp speed. All smart companies will adapt and use agents to become significantly more efficient.

Thanks, Andy McClung, CFP®

Podcast: Shell Game (Apple Podcasts) Google Search: “Shell Game Evan Ratliff”

2026 Market Results – 05/7/26