AI is Stainless Steel

When I was in high school, I took a machine shop class.

We learned how to work with steel on a milling machine and a lathe. We learned that different alloys have completely different behaviors and uses. Some are hard but brittle. Some are soft but flexible. Some resist corrosion. Some hold an edge better. The properties of the material determine what you can build with it, but the material itself isn't the product.

Why is this important now? Because everyone talks so much about AI, but AI is just stainless steel.

Nobody ever bought stainless steel. They bought a knife that doesn't rust. A pot that lasts twenty years. Surgical instruments that can be sterilized thousands of times. The steel? That's just how it works.

Right now, every software vendor on the planet is breathlessly pitching "AI-powered" solutions. They're leading with the technology like it's the product. It's not; it's BARELY even a feature!

AI isn't a solution. It's the "material." And if you're buying software based on whether it has AI in it, you're making the same mistake as someone who walks into a hardware store asking for "some stainless steel" without knowing if they need a hammer or a spoon (idiotic as that may sound).

Here's What Stainless Steel Actually Is

Stainless steel is better than regular steel. It doesn't rust. It's more durable. It's easier to maintain. Those are real advantages that transformed manufacturing.

But stainless steel isn't a product.

You can build a hammer out of stainless steel, and that hammer won't rust. Great. But the material isn't what makes it a hammer. The form, the weight, the handle, the striking surface – those are what make it useful. The steel just makes it better at being a hammer.

Now imagine if hardware stores started selling things like this: "Buy our stainless steel solution!"

You'd ask: "Is it a hammer? A wrench? What does it DO?"

That's the conversation we should be having about AI. But we're not.

The Problem: Everyone's Selling Stainless Steel

Walk into any software pitch today and you'll hear some variation of: "Our AI-powered platform leverages advanced agents to transform your workflow."

Stop. Transform it how?

"AI-powered" has become meaningless. It's this decade's "cloud-based" or "mobile-optimized" – a checkbox that everyone claims and nobody can explain the value of.

Here's the reality: AI and Large Language Models are the base material. Agents are construction techniques. But what you should care about is the actual feature and what it does for you. Not the technology behind it.

Think about how you actually use AI:

ChatGPT and Claude aren't valuable because they have AI. They're valuable because they help you research faster, write better, and solve problems. The AI is the how, not the why.

Google's AI summaries don't matter because they use AI. They matter because you get answers without clicking through ten links.

Grammarly doesn't sell "machine learning models." It sells "never send an email with typos again."

See the pattern? When AI works, you forget it's AI. You just notice your problem got solved.

The Red Flags That Should Make You Run

If you're shopping for software, here's what should set off alarms:

1. Leading with AI as the feature.
   If the first thing they tell you is "AI-powered platform" instead of what problem it solves, they're selling you stainless steel without telling you if it's a knife or a paperweight.
2. Vague promises about transformation.
   "Uses advanced AI agents to revolutionize your workflow" means nothing. What gets faster? What gets easier? What becomes possible that wasn't before? If they can't answer that, they haven't built anything real.
3. It's just ChatGPT in a trench coat.
   Could you do the same thing by pasting your data into ChatGPT? Then what are you paying for? If the answer is "well, we integrated it for you," that's not a product, that's a convenience tax.
4. No specific outcomes.
   "Improves efficiency" is not an outcome. "Reduces ticket resolution time from 3 hours to 20 minutes" is an outcome. If they won't give you numbers, they probably don't have any.
5. Demos that show the AI instead of the result.
   If they're walking you through "how the agent reasons" instead of showing you the end result, they're admitting they don't have a compelling product. You don't buy a knife to learn about metallurgy.

The Questions That Cut Through the Noise

Here's how you separate real products from technology demos:

1. "What problem does this solve?"
   If they can't answer this without saying "AI," keep walking. The technology should be invisible in the value proposition. What's ironic is that any company that has AI in it's tech stack seems to be getting huge funding. Don't be fooled by funding and hype.
2. "Show me the before and after."
   Not with their cherry-picked demo data. With your actual use case. If they can't demonstrate measurable improvement on your problem, they haven't solved it.
3. "What happens when it's wrong?"
   AI makes mistakes, and it's not deterministic in its behaviour. Good products account for this with guardrails and human review. Bad products assume the AI is magic. How they handle errors tells you if they've actually thought through production use. It's also important to look at your liability with the actions that tools built with AI are taking.
4. "Why can't I just use ChatGPT?"
   You might! ChatGPT is a great product that can do a ton of stuff, but you already know that. However, the best AI products that AREN'T ChatGPT have specialized domain expertise, proprietary data, or deep workflow integration that you can't replicate with a generic LLM. If they can't articulate this clearly, you're about to overpay for a wrapper.

What Actually Good AI Products Look Like

The best AI products don't just talk about AI.

Salesforce Einstein doesn't pitch "advanced AI predictions." It pitches "know which deals will close."

GitHub Copilot doesn't advertise its model architecture. It advertises "write code faster."

Grammarly doesn't mention machine learning. It promises "catch every error."

These products solve specific problems. They measure success in outcomes, not in "AI accuracy scores." They enable things that were impossible before – real-time translation, personalization at scale, analysis of unstructured data that humans can't process.

And here's the key: the AI should be somewhat invisible. You just notice your work got easier.

Stop Buying Technology, Start Buying Solutions

When you evaluate software, start with your problem. What are you trying to accomplish? What's the measurable outcome you need?

Then demand specificity. Ask vendors to explain their value without mentioning AI. If they can't, there's no value there. Request real case studies with actual metrics. Test on your data, not theirs.

You're not buying AI. You're buying a solution to a problem you have.

The technology is how it works. You should be evaluating what it does and why it matters.

Wait, But What About Developers?

Okay, here's the gotcha you've been waiting for.

Tool manufacturers DO buy raw stainless steel. Of course they do. Someone has to turn that alloy into hammers and knives and surgical instruments.

The same is true for AI.

Developers absolutely should be buying "raw AI" – APIs from OpenAI, Anthropic, Google, and others. That's the whole point. These APIs are the raw material that developers use to build actual products. Developers need access to the raw models in order to apply domain knowledge, MCP tools, and system prompts on top of them.

If you're a developer building agents, you SHOULD care deeply about which LLM you're using, what its capabilities are, how much it costs, and how reliable it is. That's your stainless steel supplier. You need to know the material properties because you're the one forging it into something useful.

The Anthropic API, the OpenAI API, the various model providers – those are legitimately selling you AI as the product, because you're the manufacturer. You're going to take that and build something people actually want.

But here's the critical distinction: when you turn around and sell what you built, you better not be pitching "we use Claude Sonnet 4" as if that's the value proposition. Your customers don't care. They care about what your tool does for them.

GitHub Copilot is built on LLM models, but they don't sell "LLM integration." They sell "autocomplete for your entire codebase."

Notion AI uses various LLMs under the hood, but they sell "write, edit, and summarize without leaving your workspace."

Grammarly uses AI models, but they sell "make your writing clear and mistake-free."

So yes, if you're building products, buy the raw AI. Study it. Understand it. Choose the right model for the job.

But when you're done building, sell the hammer, not the steel.

A Last Word

AI amplifies everything it touches. Build it into a real solution, and it might scale your approach in ways you never imagined. But amplification isn't improvement—it just means more output, faster. AI won't fix a broken solution; it'll just help you produce broken results at scale. Make sure you're solving the right problem before you amplify anything.

Which direction feels right to you?