Hand me a welder's torch and somebody is going to get hurt.

That isn't false modesty. I have no business holding one. Sit me at a piano and you get the same result in a different key, since my last lesson was somewhere around 1977 and whatever survived of it is long gone. Put a canvas in front of me and you'll get stick figures with an abstract-art defense already prepared.

None of that says anything about the torch, the piano, or the paint. Those are excellent tools. They are simply useless in my hands, because a tool does not carry the skill. The person does.

That is the whole argument about artificial intelligence, and most of the public conversation is having it backwards.

What competition just did to the price of AI

Start with the part of this that is pure economics, because it happened fast enough to watch in real time.

A few weeks ago I pointed out that Kimi 3.5 had shipped at roughly half the cost of Fable 5 with close to the same capability, and asked the obvious question. Why would anyone stay on the expensive option when the cheap one does nearly the same work?

Within days, Anthropic released Opus V at about half the price of its own flagship, with capability that benchmarks close behind it. I have been running Opus V on high effort for almost everything since, and in my own use I have been happier with it than I was with Fable 5 at maximum effort.

I would not read that as a story about one company being clever. Read it as market forces doing what they are supposed to do. A serious competitor showed up at half the price, and the incumbent cut its own price rather than defend the margin. Every business owner who has ever been the only vendor in a market should sit with that for a minute. Monopolies are comfortable for exactly as long as they last, and they never last as long as the monopolist expects.

The practical takeaway for you is less philosophical. The cost of capable AI is falling, quickly, and the tools you priced out eighteen months ago are worth pricing again. Whatever you concluded about this in 2025 has probably expired.

(Pricing and benchmark positions move constantly. Verify the current numbers before you budget against them.)

I built our firm's chatbot, and I cannot code

There was some confusion after my last video about an internal chatbot we've been developing, so let me clear up two things.

First, the name. Everything inside our firm is mountain-themed, because that's how we think about the work. Clients are climbing something, and we're the guide who has been up the route before. So the guide gets called a Sherpa, and our internal chatbot, which runs on my second brain, is named Sherpa.

That has nothing to do with Sherpa.ai, which is a real and separate AI platform. We don't use it, don't sponsor it, aren't sponsored by it, and I didn't know it existed until I mentioned the name in a video. If you went looking for a connection, there isn't one.

Second, and more useful to you: I built ours with Claude Code, and I have no programming background. The last time I wrote code it was Pascal, in roughly third grade, close to fifty years ago. Whatever that taught me did not survive either.

If you haven't sat down with Claude Code or one of its competitors, you should. Almost anything you can describe as a computer program, an iPhone app, an Android app, or a website is now something you can build by explaining it in plain English. You describe what you want it to do. It writes the program.

Then the part people don't expect: it walks you through getting the thing live. You ask how to deploy it, how to put it somewhere your team can reach it, how to get it onto your website, and it answers. It will package the whole thing up so you can hand it to your IT company. When they email back with technical questions you don't understand, you paste their email in, and you paste the answer back to them.

You will look considerably smarter than you are. I say that as someone actively doing it.

The three-platform problem

Here is where this stops being a novelty and starts showing up on your P&L.

Most established businesses I talk to are running at least three systems that don't speak to each other. A CRM. An accounting platform. Some kind of operating or project system. Somebody on your team spends real hours every week exporting from one, reformatting it, pasting it into another, and assembling a report that leadership looks at for ninety seconds.

That person is not doing skilled work during those hours. They are doing translation.

You can now sit down with an AI coding assistant, describe the three platforms you use, arrange proper API access, and describe the report you actually want. Every Monday morning, or every Friday evening, tell me these numbers. It will build that for you, and the finished thing can live somewhere ordinary like SharePoint or Dropbox where you click it when you want it.

One caution I want to be explicit about, because I glossed it when I first talked through this: Connect systems through proper credentials and API keys, scoped to what the tool actually needs, and involve whoever handles your security before you wire anything to your accounting system. The convenience is real. So is the exposure if you hand broad access to something you set up in an afternoon.

The second-order effect is the one worth planning around. A meaningful share of what you pay for in software subscriptions is glue: connectors, reporting add-ons, workflow tools, the middleware tax. Some of that you are about to stop buying. Not all of it, and not the systems of record themselves, but the layer between them is getting very cheap to build yourself.

We used to tell everyone that the answer was STEM education, and I don't disagree. Science and math and engineering still matter enormously. But the specific claim that programming was a specialist priesthood has quietly stopped being true. Most people can now build the particular thing they need.

Where "anyone can do anything now" falls apart

Which brings me back to the torch.

You still must have the underlying foundation in whatever you are doing to use AI to any real extent. For a welder, the tool is the torch. For a musician, the piano. For a painter, brushes and canvas and an eye. For lawyers, our tools are the statutes and the case law and twenty-five years of watching how the same fact pattern goes wrong. For a business coach, the tools are the books and the psychology and the history and the hours in the room.

We all know how to use our own tools. AI is another one on the shelf.

So when a layman sits down and asks it to produce art, or music, or a computer program, or a legal pleading, or a read on a medical symptom, the result is roughly me with the welding torch. Noise, mess, and a nonzero chance somebody gets hurt.

I was pushing this on a friend recently. He has a filmmaking background, real experience with actors and lighting and everything else that goes into the craft, and he immediately shut the conversation down. It doesn't work. The eyes are wrong. He wasn't ready to hear the point I was making, which was never that the tool is finished. It was that he is exactly the person who should be holding it.

If Steven Spielberg sat down with an AI to build a scene, the result would be better than anything I could produce by a factor I can't count. Not because he has better software. Because he has the background, the instruction, the education, and the vocabulary of the profession. He knows what he is looking at when it comes back wrong, and he knows what to ask for next. I would not.

That is the asymmetry nobody selling AI courses wants to talk about. The tool amplifies. Amplifying zero returns zero.

Use it inside your craft, where you have the experience and the education and the lexicon and the scar tissue, and it genuinely expands what you can do, because it becomes a thought partner rather than a vending machine. Point it at a field you know nothing about and it will hand you something confident, plausible, and wrong, and you will not have the equipment to notice.

Why the liberal arts crowd should relax

There's a loud version of this argument going around that says nobody needs school anymore, knowledge is a commodity, and liberal arts education is finished.

That is wrong, and I think it's wrong in a way that will cost people money.

What gets accelerated by these tools is what you already carry. Your experience, your history, your judgment, your sense of what a situation actually is. The AI multiplies that. It does not supply it. Strip out the human underneath, and there is nothing left to multiply.

Yes, my own AI now surfaces ideas I didn't feed it, drawn from what it has learned working alongside me over time. And yes, some later generation may genuinely think for itself. It isn't there today. Today it's the paintbrush in the artist's hand and the torch in the welder's.

I have a self-playing piano at the house. It will play Elton John, Billy Joel, and Adele on demand, and it's a pleasure to have. It is also not remotely the same as any of those three sitting down at it. Their fingertips and their feeling put something into the music that the instrument by itself will never produce.

Getting output is not the same as having the skill. Asking a question and receiving an answer does not make you an artist, a welder, a musician, or an expert in a field you've never worked in, and it does not mean you no longer need the people who are.

What to do with this on Monday

Two moves, and they run in opposite directions.

1. The first is to stop treating custom software as something you buy. Pick the single most annoying manual process in your business, the one where somebody moves data between systems by hand every week, and spend an afternoon describing it to an AI coding assistant. Loop in your IT support before anything touches live data. You will either end up with a working tool or with a very clear understanding of why it won't work, and both are worth the afternoon.

2. The second is to stop pretending the tool is the expertise. If you're using AI inside your own profession, push it hard, because that's where it pays. If you're using it to produce work in a field you don't actually know, you're the one holding the torch, and confidence is not the same thing as competence.

So which one are you doing right now, and which one do you think you're doing?