AI expertise is not one thing. Learn how to distinguish deep experience from shallow credentials, hype, and expertise that does not match the problem you need solved.
These days everyone wants to hire people who are experienced with AI. Fortunately, there are now literally millions of people who claim to be AI experts. Unfortunately, there are now literally millions of people who claim to be experts. Before you hire such an expert, take a moment to understand just what “expertise” means. Most are self-proclaimed, and even those with third-party credentials may be wearing clothes that, while not quite invisible, are pretty threadbare.
Deep expertise takes years of learning and experience. Today’s gold rush has created a cottage industry of low-value expertise gained in a matter of hours. How can you know who is a true AI expert and who is cosplaying as one? You need to do your homework and look deeper at their expertise.
Consider the term "automotive expert” (something I most certainly am not). That might be a great auto mechanic, one who can fix anything. That’s different from someone who is an expert at designing cars. Certainly an automotive engineer who designs cars probably knows a thing or two about fixing them, but doesn’t spend all day working below the undercarriage of a car. Likewise, the auto mechanic may have some knowledgeable opinions on car design, but doesn’t spend his days thinking through design tradeoffs. Then there are people who write for automotive magazines and websites. They have extensive knowledge of some combination of engineering, sales, trends, or industry knowledge. Finally, there are the people who you may turn to when it comes time to buy a car; e.g., your friend who you know goes to auto shows and reads up on things. It’s not his profession, but he’s more knowledgeable than the average person. Again, people in one category may have informed opinions about the others, but that does not make them deep experts outside their own area.
All can be considered experts, but their expertise is not interchangeable, and depending on your needs one can be patently wrong. Of course, to those with no knowledge of cars at all, any can seem knowledgeable. The problem is, when hiring for a role, you need to have some minimal knowledge of what is needed in order to evaluate it. Consider, if you had to hire a commander of a lunar mission, what would you be looking for? Unless you’ve been working at NASA, you probably have no idea and would simply be guessing at what’s important.
The same is true in AI. First I’m going to narrow the field. AI has many specialties including machine learning, computer vision, expert systems, etc. They are deep in their areas, but not necessarily experts on generative AI, which is what we’ll focus on since that’s what most people are doing these days. As an analogy, consider someone with a PhD in US colonial history; how much do they know about classical Greece? They might know a little more than the average person since their education included history broadly, and they could probably pick it up faster than the average person, but it’s really not their area.
For the rest of this article, AI will refer to only generative AI (although similar analogies could apply to the types of AI, too). There are the model builders, solution builders, domain practitioners, and educators/speakers.
The model builders are the PhDs and equivalents who create the foundational LLMs. They typically work for companies like Google, Anthropic, and OpenAI. Like their automotive counterparts, they work with experts in product design, QA, safety, and related engineering fields. Just as you don’t need an automotive engineer unless you’re designing a car or similar vehicle, unless you’re creating your own LLM from scratch (or similar), you probably don’t need this expertise.
The solution builders are the “mechanics.” They create AI solutions for your customers or for internal use. These may be your software engineers or data scientists who are building AI solutions and tools. While that work typically meant MCPs and RAG the past few years, today it’s more likely to include agentic AI.
As the tooling rapidly matures, there’s a low barrier to entry here. You don't need to be a software engineer to use many of the tools. This is a double-edged sword. On the one hand, this is one of the reasons “expertise” has become so commonplace. I’ve seen people create simple wrappers around commercial LLMs, the kind that can be built in as little as thirty minutes, and call themselves an expert because in a week they built a few dozen of them. The reality is much of this expertise is readily accessible to non-experts who invest a few dozen hours to learn. This is made all the easier by all the blogs, videos, and even support you can get from an LLM.
On the other hand, people who understand a domain well and go deep in applying AI tools to problems specific to that domain or functional role do start to gain expertise. Remember that most roles have an 80/20 rule. The base case is straightforward; the expertise matters when dealing with the exceptional cases. Someone who understands how to get the AI tool to handle those edge cases, because of her combination of domain knowledge and AI tool knowledge, has defensible expertise. Someone who simply read a blog and followed the setup instructions for a tool, not so much.
Speaking of blogs, there are also the alleged experts who write and speak about AI. If you post enough in your blog or social media, you start to look like an expert, whether or not you are one. I have one foot in the professional speaking world and watched as every speaker and their brother became an AI “expert” in the past three years. Some bill themselves as general AI experts; these are typically the “leadership” and “futurist” speakers. Others had their industry niche (e.g., real estate, beauty industry, aviation) and became an “expert” on AI in that industry.
Be careful of these so-called experts since most of them haven’t really used AI, or if they have, they built something in a week as evidence of expertise. The reality is, these folks who talk, but don’t do, often lack the depth to understand secondary effects and how choices will impact you one or two years down the road. To be fair, some are actual practitioners who work with AI in their day jobs and happen to write or speak about it.
For the record, I’m in two camps. In my day job as a CTO, I build with AI. That means AI-driven development, using AI in solutions, and using AI for internal work operations. (I also have a number of AI patents.) Even then, my expertise varies by industry (e.g., I don’t know much about aviation and maybe there are wrinkles for AI in aviation someone with my background wouldn’t know); my areas are general enterprise software, cybersecurity, medical, and financial services. I’m also a writer (as you know from reading my articles) about technology and professional development, and I speak around the world. Interestingly, my AI speaking came from a unique convergence. I spoke about technology, including more traditional AI/ML (based on my work as a CTO), and I also spoke in the completely separate world of professional development (based on my teaching at MIT and my book The Career Toolkit: Essential Skills for Success That No One Taught You). The two came together a few years ago when everyone started asking questions like: What does AI mean for my career? How do I manage with AI? How does AI impact hiring?
One way to spot experts is by their credentials, but that only works if the credentials themselves have value. For example, given my computer science degrees from MIT, I must know something about software. A doctor who is board-certified in dermatology is presumably more of an expert on skincare than a general practitioner.
However, not all certifications are equal. In software, the code academies (also known as code camps) famously promised to fast-track people into high-paying software engineering jobs. The reality is the people who graduated from the schools were thinly educated and most serious software companies wouldn’t hire them until they had a few years of experience. The true value of a credential varies greatly depending on how much knowledge and training underpin that credential.
There are now lots of credentials for AI like the Google AI Professional Certificate on Coursera. Be careful with these. The Google certification is an eight-hour online class for which over one million people have already enrolled. If the difference between you and an expert is an eight-hour online class, you don’t need the expert, you need a few hours to learn it yourself.
Even online classes that look more robust aren’t always. IBM’s RAG and Agentic AI on Coursera is listed as an online course lasting 3-6 months, but in reality the description reads “8 weeks to complete at 3 hours a week.” That’s just 24 hours, or a solid weekend. On the other hand, the Coursera Microsoft AI & ML Engineering Professional Certificate reads “6 months at 7 hours a week.” That’s 182 hours, or the equivalent of a full-semester class. (I’m using time as a proxy of quality, since evaluating the quality of the content takes more work. I’m assuming content quality is the same in all cases since no one has a secret horde of knowledge unavailable to others.)
Three AI certifications from Coursera. Three brand names: Google, IBM, Microsoft. But what won’t be evident on the resume is that they represent 8, 24, and 182 hours of training. There’s a world of difference in the amount of training between the first two and the third.
There’s a second problem. Just as every company is branding its products as “AI” (whether or not they use AI), every teaching institution wants to tap into the demand. The easiest way to do that is to take existing content, sprinkle in a little AI, and slap an AI label on it.
Consider MIT’s AI-Driven Leadership; much of it is rehashed content taken from general leadership and digital transformation. I’m not saying that content isn’t valuable, but it’s also not as cutting-edge AI as you might think. I knew an accounting professor at a top business school who talked about changing his class names from something like “Accounting & Tax” to “Tax & Business Strategy” because it was now viewed as a strategy class, even though the content was the same. (For the record, I chose the MIT class because of my own affiliation with MIT; I could have picked almost any class and it would have been the same, 90% prior content packaged up with an “AI” bow.)
The search for “artificial intelligence” on Coursera returns thousands of results including “AI for Executives” from Khalifa University, “Governing Data in the Intelligence Age,” and “Build AI App with ChatGPT, Dal-E, and GPT-4.” They range from a few hours to months of work. The title alone doesn’t tell you how deep the content was or how closely it focused on the AI expertise you need, even when it comes from a brand name. It’s hard to find the signal in all the noise.
With literally trillions of dollars being thrown at every level of the AI stack, there is a huge amount of money to be made up and down the value chain. It’s been said that in a gold rush, the best way to make money is to sell shovels. Expertise and credentialing are classic shovels: “I’ll tell you how to find gold,” and “I’ll certify you as a gold-finding expert so others will back you.”
In the age of social media, everyone is a talking head. Each new trend brings new opportunities. The joke in software is that every few years there’s something new a company wants: mobile, cloud, big data (machine learning), blockchain, and now generative AI. Whenever the new wave hits, companies throw big money at trying to solve it, and would-be experts line up quickly to take the cash.
Caveat emptor.
The folks with the PhDs have high-barrier expertise. They’re also probably not the ones you need unless you’re building foundational AI. The learning curve for most AI tools is pretty shallow, and if that’s what you seek, you can likely learn it rather than try to buy it; deep applied expertise is a much rarer commodity. A good software engineer without prior AI experience can learn the new AI tools pretty quickly; what an “AI-only” expert may lack is the software engineer’s years of experience with edge cases and real-world issues. A good operations professional or salesperson, or really someone in any other role, can spend a couple evenings online and learn how to use AI in their jobs. As for the speakers, if you’re hiring them because they fill an hour in an entertaining way, sure; but if you want expertise, look at whether the speaker you hire has not only built the type of AI solution you need but stuck around to see the consequences of trade-offs six or twelve months down the line. So many talk a good game, but unfortunately, talking is all they know how to do.
It’s critical to learn about corporate culture before you accept a job offer but it can be awkward to raise such questions. Learn what to ask and how to ask it to avoid landing yourself in a bad situation.
Investing just a few hours per year will help you focus and advance in your career.
Groups with a high barrier to entry and high trust are often the most valuable groups to join.
