Minds, Extended

Our tools become part of us

Pick up a rake and use it for a while. You may begin to judge where its tip is without looking, as if it extends your reach. Research on macaques using a rake found changes in how their brains represented the tool and nearby space. It is one example of how using a tool can change our relationship with it.

The same thing happens with the mind. Philosophers call it the extended mind: a tool outside your head, like a notebook, can become part of how you think. AI works the same way. It isn’t a replacement for you. It’s an extension of you. The more you use it and shape it to your work, the less you think about the tool and the more you think about the job.

Everyone gets an assistant. Everyone becomes a manager.

Everyone now has a tireless assistant, researcher, and analyst. That turns every job into a manager’s job:

1. Decide what you want.
2. Explain it clearly.
3. Give the context.
4. Hand off the work.
5. Check what comes back.
6. Correct it.
7. Make the final call.

In these tools, the language engine is an LLM. Models designed for harder multi-step reasoning are often called LRMs. An agent combines a model with tools to choose and carry out steps; its harness manages context, permissions and execution. A reusable skill supplies instructions and examples. MCP connects compatible apps to data and tools. We choose the parts needed for the job and check the result.

The core skill isn’t prompting. It’s management.

Good managers know the work well enough to spot when something has gone wrong. The same holds with AI. Your own knowledge of the job is how you catch its mistakes, so handing work to AI is no reason to stop learning your field.

Judgment: check before you trust

AI writes ten believable answers in seconds. The hard part is telling which one is right, so checking answers now matters more than producing them.

AI also sounds just as sure when it’s wrong as when it’s right. So do people: research on expert intuition shows that feeling confident is no proof you’re right. What catches mistakes is knowing your field and checking the work. With AI, critical thinking comes down to six everyday questions:

1. Did the answer address what I actually asked?
2. What is it assuming?
3. Do the sources really support the claim?
4. What’s missing, and what would change the conclusion?
5. Could I explain to someone else why I trust this?
6. Did I word the question to get the answer I wanted?

Taste matters more, too. By taste I mean knowing good work when you see it. When drafts are cheap, that’s worth more, and it improves with practice: study good work, compare versions side by side, and learn to say why one is better.

AI can help you practice. Ask it to question your assumptions, argue the other side, or find the gaps in your reasoning. Just ask honest questions. A leading question gets the answer it was looking for.

Give people back their attention

With a small business, I start with a simple exercise: list everything your organization does, then ask which tasks still need a person.

Hand what you can to AI, and check what comes back. For account permissions and data security, we work within your organization’s IT rules.

The bigger win is using the time you save for the human work that grows your business:

• understanding customers
• solving unusual problems
• improving the product or service
• building relationships
• making judgment calls

Automation is only half of it. The other half is deciding where the freed-up attention goes.

Some skills gain value, others lose it

Picture a farmer who has spent years walking behind a plow, pushing and steering it by hand. When he trades it for a tractor, he loses some strength: the machine does the heavy work now, so his arms and legs no longer get that workout every day. But he gains far more: hours back in his day, more land than he could ever work by hand, and bigger harvests that feed more people. In 1830, growing 100 bushels of wheat took 250 to 300 hours of labor. By 1987, it took 3 hours. Nobody wants the plow back.

AI is doing the same thing to thinking work. Here is how I believe skills will shift over the next five years:

Employers see the same shift. In the World Economic Forum’s 2025 survey of more than 1,000 employers, creative thinking, curiosity and lifelong learning, and leadership and social influence rank among the fastest-rising skills.

Talk, don’t type

The biggest shift is in the interface between people and computers: ordinary language is becoming a way to tell software what we want done. We can explain a goal, give context, and refine the result through conversation. I usually do that by voice. The system still needs clear instructions, and we still have to check its work.

Dictating a message is about three times faster than typing it, and talking comes far more naturally to people. Speech goes back hundreds of thousands of years. Writing is about 5,200 years old, and as recently as 1820, only about 12% of the world’s adults could read. For decades, computers made us adapt to them. Now they listen, talk, and see. They’re adapting to us.

That’s why my sessions are built on dictated prompts. You talk, the AI drafts, and you spend your time reading and judging what comes back. Reading gains value because AI produces more text than ever, and someone has to check it.

We are in the Centaur Age

In May 1997, IBM’s supercomputer Deep Blue beat world chess champion Garry Kasparov. Much of the world saw it as the end of human chess.

Kasparov saw an opening. In 1998, he played the first match of Advanced Chess, where each player had a computer at their side. In 2005, an online tournament opened the idea to anyone: humans, computers, or teams of both. The overall winners were a team: two amateurs running three ordinary PCs. They beat grandmasters with better hardware because they were better at coaching their machines. Kasparov’s conclusion: a weaker player with a better process beat both the strongest computer and stronger players with a worse process.

Chess players called these teams centaurs: human plus machine, stronger than either one alone. That’s the age we’re in with AI, and it’s good news for a small business. The winners didn’t have the best equipment. They had the best process.

Right now, people and AI are good at different things.

AI can carry out a direction. A person still has to set it. The most effective AI users I know aren’t using it to replace themselves. They use it to do far more of what they already wanted to do: research more, build more, learn faster, and take on projects that once needed more time or more people. The person doesn’t disappear. Their reach grows.

Where this is going

In chess, the centaur era was short. Human-plus-computer teams won the open tournaments from 2005 to 2008, but chess programs kept improving, and today human strategy adds little. AI may follow the same path. Nobody knows how long the centaur age will last.

I’m very concerned about where AI is going. Machines keep improving, and nothing guarantees we’ll stay in control if we build intelligence that exceeds our own. The stakes are bigger than any business.

I believe we have an obligation to pay attention and learn firsthand what these tools can do and where they fail. People who understand AI are better equipped to question how it’s used and to help shape the society we live in. Understanding these tools won’t settle every question about how AI should be built and governed. But it’s where good decisions start, at work and in our communities.

Where I come in

AI is improving faster than you can keep up with. That’s no failure on your part. You have a job, a business, a family. You don’t have time to track every model release, test every new feature, or relearn best practices that shift every few months. The result is a growing gap between what these tools can do and how you actually use them. Right now, that gap holds people back more than the limits of the technology itself.

Closing that gap is my job. I follow AI tools closely, testing and comparing them, finding what works and where they fail, so you don’t have to.

I’ve been a teacher since 2010, most recently teaching GED math in Spanish at Central Oregon Community College. I teach AI the same way I taught math: by doing. I don’t build custom AI software. We work together over Zoom, in English or Spanish. We set up your AI tools and work through real tasks from your job: email, meeting notes, reports, moving data between systems. I work the problem in front of you first. Then you take the controls and practice while I coach. I even wrote this essay the way I teach: by talking it through with AI, then checking and editing every line.

Along the way, I teach two things:

• Tool fluency: what each tool can do, how to give it context, how to work with files, voice, and agents, and where it fails.
• Judgment: when to trust the output, how to check it, and how to supervise the work.

Tool fluency without judgment produces people who can run the machine but don’t know what to ask it or when to believe it. I teach both.

Experiment freely. Trying things and making mistakes is how anyone learns a new technology, but nobody needs to repeat every mistake the people before them already made.

The goal is independence. The technology moves fast, so periodic check-ins help, but success means you need me less over time.

Small businesses are my main clients. Schools face the same shift: when AI can explain and give practice with endless patience, a teacher’s time can go to mentoring, motivating, and noticing who is struggling. That’s why I also offer this training free to educators, nonprofits, and community organizations. Useful AI skills should reach everyone, not only the people with the time and technical confidence to explore on their own.