Learning AI

Many of us are not sure how to use AI. This series of posts is intended to show you the different ways AI can augment your intellectual capabilities. Whether it’s a simple question or a more complex challenge, we will test AI and see how it can work to give you a problem-solving advantage.

A diverse set of topics will be covered, but most importantly, we will publish the actual AI dialogue so you can see how it all works.

Each week, we will select two of the AI engines and pose a question or a problem, summarize the findings in our own words, and include the interactions with the AI for your review.

We posed the following question to ChatGPT and Google Gemini:

What’s the difference between a chatbot and an agent?

Before going into the responses for this week’s question, let’s first understand how Google Search works, so we can differentiate it from AI chatbots and agents.

Think of Search as an indexing system. All day and all night, Google’s computers scan every page added to the web, identify keywords on the page, and store the keyword information about the page on Google’s computers. When you search for great weekly newsletters, TheCautionary.com shows up on the search list since it was identified as a great newsletter by the Google indexing engine.

Your search (if you are willing to go to page 32 of the search results), presents a link to this newsletter that when clicked, directs you to the site. Google Search does not store the website; it directs you to it.

Moving to chatbots and agents, here is a quote from one of the AI engines to help understand the difference between a chatbot and an agent:

As long as you are analyzing a known set of information and not the entire Internet, chatbots are great. For example, organizations put chatbots on top of things like technical manuals and shopping sites.

With a technical manual, the chatbot consumes it all into its memory and storage so that when you ask the chatbot what kind of oil to use for the engine, it instantly retrieves the information from the manual and displays it for you. Same with a shopping site; the chatbot can go find the cheapest umbrella in the inventory and show it to you.

Depending on the site, sometimes the chatbot is programmed with a little personality and might say, “Dude, that was easy, throw something tough at me.”

Agents are a whole new ballgame and a game to be played carefully.

As AI tools have advanced, much focus has been on personal productivity. Let’s use travel as an example. To take a trip for a long weekend, you need to book flights, rent a car, and reserve a hotel. That’s several websites or a lot of annoying time on the phone.

A properly designed agent can do this for you. But, and please remember this, you must invest the one-time work to set up the agent with your permissions and preferences.

The first permission you give the agent is access to your calendar. Decide if the access is read-only or if the agent has permission to write to your calendar or reschedule appointments (see what I mean about being careful?).

Next, the agent needs your permission and login credentials for the airline, car rental, and hotel sites. You can tell the agent that you want a window seat on all flights, a convertible rental car, and a hotel that folds the ends of the toilet paper into those fancy triangles.

Most of the set up for an agent is one-time with updates like adding new hotel permissions along the way. Agents are powerful tools, but need to be used with caution, else you may end up with an agent that booked thousands of dollars of non-refundable reservations at roadside motels with shared bathrooms and scenic views of the county dump.

The AIs also told us that chatbots generally wait for the next question or response from you while an agent goes out and does things on your behalf (once the agent has proper permissions.)  

Unlike chatbots that regurgitate information, agents can summarize today’s news, analyze a spreadsheet, or do a web search for stocks that are experiencing abnormal price or volume increases.

Chatbots are good tools that most of have used. That awkward thing on Amazon called Rufus? That’s a chatbot. 

Agents are the next wave of technology starting to wash over us. Do you need to be an expert to use an agent? No, but there is value in understanding how AI can do more than help with that oil change.

We will walk through building an agent in a future post. If you just can’t wait, play around on the following sites. They are designed for non-programmers who want to build agents and both offer a free subscription level. Email me if you get stuck.

Base44.com

Lovable.dev

Things I think about

The first computer bug was an actual bug. It was a moth found trapped in a relay circuit of the Harvard Mark II computer in 1947.

Recommended Reading

The Reckoning
David Halberstam's account of the American auto industry's decline.

The Four
The hidden DNA of Amazon, Facebook, Apple, and Google.

Hard Drive
Bill Gates and the making of Microsoft.

No Filter
The inside story of Instagram's founders and their rise to fame.

Response from Google Gemini

What’s the difference between a chatbot and an agent?In 2026, the distinction between a chatbot and an AI agent has become a fundamental shift from conversation to action.

Think of a chatbot as a knowledgeable librarian you can talk to, while an AI agent is a capable intern you can delegate work to.

Key Differences at a Glance

1. The "Logic Loop" (Reasoning)

A chatbot follows a linear path: You ask, it answers. Even the most advanced chatbots in 2026 generally wait for the next prompt before continuing.

An AI Agent uses a reasoning loop. If you tell an agent to "organize a team dinner," it doesn't just give you a list of restaurants. It will:

1. Check your team's shared calendar for availability.

2. Cross-reference dietary restrictions from past emails.

3. Search for highly-rated spots nearby.

4. Self-Correct: If the first choice is booked, it goes back to step 3 without asking you.

2. Multi-System Integration

Chatbots usually live within one app (like a website's help bubble).

Agents operate across your entire digital "stack." A modern agentic system can pull data from a PDF in your Google Drive, summarize it into a Slack message, and then create a task in Asana based on the results—all from a single initial command.

3. Memory and Adaptability

• Chatbots have "short-term memory" (the current conversation).

• Agents have "long-term context." In 2026, personal agents remember your preferences over months. If you previously told your agent you hate flying on small planes, it will automatically filter those out of future travel planning without being reminded.

Real-World Comparison: The "Flight Cancellation" Scenario

• Chatbot: You ask, "My flight was canceled, what are my rights?" The chatbot provides a summary of the airline's policy and a link to the refund form. You still have to fill out the form and find a new flight.

• AI Agent: You say, "My flight was canceled. Fix it." The agent identifies the cancellation via your email, rebooks you on the next available flight, updates your calendar, notifies your hotel of the late arrival, and files the refund claim for the original flight.

Would you like me to try acting as an agent for you—for example, by researching a specific task and drafting a plan of action?

Response from ChatGPT

What’s the difference between a chatbot and an agent?The short version:

A chatbot talks.
An agent acts.

Both use AI, but their purpose and capabilities are different.

Chatbot

A chatbot is designed mainly for conversation.

What it does

• Answers questions

• Explains concepts

• Writes or edits text

• Provides recommendations

• Simulates dialogue

Key traits

• Reactive – it responds when you type something.

• Text-focused – output is usually words.

• No independent goals – it doesn’t “decide” to do things on its own.

• Limited external action – it usually can’t change files, send emails, or place orders unless specially connected.

Example

You ask:

The chatbot reads it and gives you a summary.
It doesn’t go do anything in the world beyond generating text.

Agent

An agent is designed for taking actions to achieve goals, not just talking.

What it does

• Can plan multi-step tasks

• Uses tools (web search, calendars, databases, APIs)

• Executes actions (book meetings, update spreadsheets, run code)

• May operate with some autonomy

Key traits

• Goal-oriented – you give it an objective, not just a question.

• Proactive – it can decide the next step.

• Tool-using – it interacts with software or systems.

• Stateful – it often tracks progress over time.

Example

You say:

An agent might:

1. Search flight sites

2. Compare prices

3. Ask you to confirm

4. Complete the booking

It does things, not just describes them.

Side-by-Side Comparison

Important Overlap

Modern systems often blend both:

• A chatbot can feel agent-like if it has tools.

• An agent usually still chats to clarify instructions.

So the distinction isn’t absolute — it’s about emphasis:

• If the system’s main value is communication, it’s a chatbot.

• If the main value is accomplishing tasks, it’s an agent.

Using the OpenAI interface like I am now, is this a chatbot or an agent

Using the OpenAI interface the way you are right now, you are primarily interacting with a chatbot — with some agent-like capabilities available in certain situations.

Why it’s Mainly a Chatbot

In this interface:

• It responds when you send a message (reactive).

• Its main output is conversation and text.

• It does not set its own goals.

• It does not independently go off and do things without your prompt.

So by default, it behaves like a chat-oriented assistant.

Where It Starts to Look Like an Agent

It becomes agent-like when it can:

• Use tools (web browsing, code execution, file analysis, scheduling tasks)

• Perform multi-step actions on your behalf

• Work toward a defined objective instead of just answering one question

For example:

• “Analyze this spreadsheet and create a chart.”

• “Search the web and summarize today’s news.”

• “Remind me every Friday to send a report.”

In those moments, it’s acting more like an agent, because it is doing tasks, not just talking.

The Best Way to Think About It

Default mode: Chatbot
Extended mode with tools and tasks: Agent-like assistant

It’s not a fully autonomous agent because:

• It doesn’t initiate actions on its own.

• It doesn’t have independent long-term goals.

• It only acts within the boundaries you set.

So the most accurate label is:

**********