AI is Snake Oil

A brief history of problems, solutions, and fraud in the wild, wild west.

I want to tell you a story about a new technological era…

This new technological era brought us new tools that can connect people and commerce in ways that were unimaginable just a few decades prior. Organizations are seeing the potential to span entire industries and geographies at unprecedented speed, shifting how we think about distance, communication, and productivity.

This shift altered the landscape of business and work itself, creating opportunities to build systems that run faster, operate more flexibly, and bridge gaps that once felt insurmountable. With these new tools, we have witnessed a redefinition of boundaries. The typical constraints—of distance, time, and even workforce limits—are being re-engineered, allowing people and businesses to reach farther, work smarter, and think bigger.

I’m not talking about Generative AI. I’m talking about the first Transcontinental Railroad, built in the 1860s.

This was a giant inflection point in American history. Connecting both coasts via rail enabled massive economic expansion, but it also ushered the United States into another new era… The era of snake oil.

A Brief History of Snake Oil

In the 1860s, a group of mostly Chinese immigrants labored for 12+ hours per day constructing what was, at the time, the longest contiguous railroad line in the world… The American Transcontinental Railroad. It was a remarkable feat, connecting the east and west coasts of the United States and allowing for greater mobility and economic opportunity for Americans from sea to shining sea.

But this labor was back-breaking, and the laborers responsible found themselves needing a salve. They had a remedy that they brought from their homeland… Snake oil, a medicine used to treat arthritis and bursitis. It was an oil derived from the Chinese water snake, which is rich in omega-3 fatty acids and has anti-inflammatory properties. They applied this oil onto their inflamed joints and saw real, measurable results that have since been validated by modern science as effective.

So, why do we remember snake oil as fraudulent quackery instead of a useful remedy from traditional Chinese medicine? The answer is a man with two first names (red flag), Clark Stanely.

Clark Stanley, the self-proclaimed “Rattlesnake King,” rose to prominence in the early 1900s selling his “so-called snake oil” as a treatment for joint pain and rheumatism. He claimed that his snake oil was made from rattlesnakes, even doing live demonstrations where he killed a live rattlesnake on stage to demonstrate how the oil was made.

Unfortunately, rattlesnakes are far less beneficial than Chinese water snakes for health problems relating to inflammation due to their low fatty acid content. Rattlesnake oil would hold none of the same healing properties of water snake oil, but was being sold as a cure-all with even more healing properties than its counterpart. The reality of Clark’s fraud is even more grim than that. As it turns out, his products contained no actual snake oil at all—just mineral oil, beef fat, red pepper and turpentine. Yet he got away with deceiving his customers for more than two decades.

So, what can we learn from the origins of snake oil and how can we use this story to better-inform our approach towards the brave new world of generative AI?

The Parallels Between Snake Oil and AI

Snake oil—the stuff made from Chinese water snakes, not rattlesnakes or turpentine, has a few characteristics that are highly relevant to the world we find ourselves in today…

• It’s useful to solve a specific problem

• What you put into it really matters

• A lot of people are going to try to get rich selling it as a cure-all for your ails

Contrary to popular belief, snake oil is NOT a scam! The same can be said about the current trough of disillusionment that we find ourselves in regarding AI.

There are a lot of techno-pessimists out there bemoaning everything that AI can’t do. This is a limiting belief that will bite a lot of people in the end.

Yes, AI is bad at a lot of things… Snake oil can’t cure cancer and today’s AI can’t replace humans at a lot of tasks. That doesn’t make either one useless. It’s our job as the users of these tools to understand how to leverage them effectively. So, let’s go through the three key parallels between AI and snake oil to break down what a good implementation of AI looks like and how you can avoid buying shelfware that doesn’t advance your business forward.

It’s useful to solve a specific problem

The first question to consider in any AI implementation is “what specific problem do I want this to solve?”

This question sounds easier to answer than it actually is, because we need to not only specify a use case, but also think critically about the underlying fundamentals of the problem we’re trying to solve.

One of the key missteps we saw in the SaaS-bloat era was an additive bias where we assumed that adding single-point solutions was an inherent good. If I don’t have an ABM software platform, I need to buy a software platform to do ABM. This circular logic leads people to over-scope their needs and under-scope the process necessary to actually implement a solution. Let’s break down some examples to see the difference…

Solutions AI is Good At, And How You Should Scope The Problem

• Content, code, and idea generation

  • Bad example: I want to use AI to write blogs for us.

  • Better example: Today, our maximum throughput for new content is 2 blogs per week. I believe that, by leveraging AI, we can increase that throughput to 4 blogs per week without degrading quality. This should lead to an increase of at least 25% to our inbound lead flow over 6 months.

• Information extraction and summarization

  • Bad example: I want AI to summarize meeting notes for all of my meetings.

  • Better example: Today, the vast majority of our meetings result in no documentation that is shared universally with our teams. I want to extract key takeaways from these 3 key team meetings, summarize them, and share them with our entire go-to-market function without putting in a lot of work. This should increase visibility on key initiatives and reduce confusion as teams work to execute.

• Interactive role playing, such as for sales calls or negotiations

  • Bad example: I want my sales team to use AI to practice cold calls.

  • Better example: The current limiting factor of my sales team practicing cold calls is the availability of a well-trained partner. I want my sales team to practice at least 3x per week with different personas and testing out various objections to identify new areas of opportunity and refine their craft. This should result in a 10% increase of connect > booked meetings over the next year.

This is the level of depth we should have when we’re standing up a new AI solution. If you go to AI looking for a generic cure to an unspecific problem, you won’t know if it’s working and won’t be able to determine fact from fiction. This is how you end up spending years rubbing turpentine on your bum knee.

What you put into it really matters

Generic LLMs are trained on a mountain of training data that absolutely everyone in the world has access to. There is nothing that you can do with a stock LLM that is unique to your business. If you try, you’re betting on your ability to out-prompt your competition forever, which is a terrible game to play.

The real magic of LLMs comes from the training data you provide them that no one else in the world can provide. How you capture, organize, and refine this data is essential to effectively leveraging AI against it. AI experts laud these models for their ability to handle unstructured data… and they can… but they’re far better at handling structured data, so why make it harder on the poor robot?

That said, many organizations don’t really know where or how to begin to organize their data to train an LLM. That’s why so many are trying to have models ingest their entire Google Drive and then wondering why the outputs from the LLM aren’t very good.

My perspective, which I may be wrong about, is that for most tasks in most businesses, I’d rather train an LLM on a small batch of well-curated data than a large batch of unstructured data. More isn’t always better. So, where do you begin?

Here are some examples of the training data I’d focus on if I was starting from square one:

• Internal knowledge base: Your internal knowledge base is going to be as important to the next decade as your CRM was to the last decade. How your organization saves, shares, and structures internal knowledge is the single highest-leverage opportunity in your business, and for most it’s a complete afterthought. If I was looking to leverage AI in any aspect of my business, the first place I’d start is with a well-structured internal knowledge base. If you have additional questions on this, I think it could be an entire standalone post. Let me know in the comments below.

• Templates and examples of what “good” looks like: In addition to a robust internal knowledge base, I’d train these models on what a good output looks like with as many examples as possible. This will both improve its output, and give you a wide open lane to learn from AI about ways to improve your own processes. For example, I may upload 10 examples of content strategy documents and ask the model to tell me areas that those documents could be improved, then refine those examples even further before I add it as training data.

Just like snake oil, what you put into AI matters a lot. If it’s made of great internal data, it could cure some real pain. if it’s made out of stock internet bullshit, it’s useless.

A lot of people are going to try to get rich selling it as a cure-all for your ails

My biggest frustration with the AI hype cycle is that far too many people over-estimate what AI will be capable of doing in 5 years while under-estimating what they could accomplish with AI in 5 months.

AI is not a panacea. It won’t magically solve all your problems. Hell, unless your problems are pretty damn small, it won’t magically solve any of your problems.

AI is a tool that requires skill and hard work to leverage effectively. If you’re not putting in the hard work to develop these skills, habits, and mental models necessary to thrive in an AI-powered world, you’re not ready to shop for solutions.

We’re seeing a troubling increase in fraud in the AI space: untested tools that promise miraculous capabilities, vendors claiming they’ve built “autonomous” solutions that don’t need oversight, or flashy models with little transparency about how they actually work. The incentives to overstate capabilities are high because businesses are desperate to get ahead and will invest heavily in the promise of AI even when it’s not well understood.

The best defense against this flood of AI snake oil is education and discernment. In the same way medical institutions eventually declared snake oil fraudulent, today we need authoritative voices to set realistic expectations and call out questionable claims. Look for established experts, read the fine print, and don’t take grand claims at face value. AI’s real value comes from its correct application in well-defined use cases. Staying informed about what AI can and can’t do isn’t just practical—it’s a necessary step in ensuring we get the best out of this technology, without falling for the hype.

P.S. - If you’re looking for AI training and consulting to level-up your team, I highly recommend Pragmatico. This group of founders has the combination of AI expertise and practical business know-how to ensure you can implement AI effectively in your organization.

How Can You Avoid Being Bamboozled?

AI’s potential is real, but so is the potential for misuse, and with so many vendors and influencers trying to sell the latest “AI solution,” it's easy to get swept up in flashy promises. The reality is that successful AI adoption isn't about finding the fanciest tools; it’s about clarity, intentionality, and a razor-sharp focus on outcomes. Companies that avoid the hype and focus on defined, measurable results are the ones that will actually benefit from AI—rather than falling victim to bloat, inefficiency, or outright fraud.

With this in mind, let’s break down a few straightforward strategies to keep you grounded and help you maximize the real value of AI in your business. These tips aren’t about quick wins or miraculous promises—they’re about building an AI approach that’s sustainable, effective, and truly tailored to what your organization needs.

• Shop for use cases, not for tools: The more specific you get about the business outcomes you want this technology to drive, the more likely you’ll achieve them. Don’t buy a vertical-specific or use-specific solution without an ABUNDANCE of clarity on the business outcome you’ll drive with this change and how you’ll measure its efficacy.

• Treat your own data like oil being refined into gasoline: The real value in generative AI isn’t in the output, it’s in the input. Your company’s internal data, knowledge base, perspective, and lessons learned is the ONLY thing left that can’t be copied with a click of a button. Treat this data like the liquid gold that you refine into fuel for your business.

• “Dream on, but don’t imagine they’ll all come true”: This classic Billy Joel line hammers home a simple truth about emerging technology… We all have a vision of where thing could go. Some of us see a dystopia, some of us see a utopia, and a lot of us see something in between. Regardless of what camp you fall into, imagining the future is an essential skill this decade, but being deluded by the future is a big, big weakness. Avoid it at all costs.

For many of us in tech, the playbook is officially broken. We’re back to first principle thinking and throwing out the old way of doing things. We’re back into academic debates about definitions and limitations and out of the “get rich quick” bubble of ZIRP. None of this is easy, but all of this is good. It’s good for our society, our industry, and the muscle sitting between our ears. The way forward is to get smarter. Let’s go get it.

What questions do you have about AI and the future of work? Comment on this post or send me an email at timhickle@gmail.com. I just may feature your question in a future article.