“What types of people do I need to hire to build AI solutions for our company?” I've been asked this question repeatedly over the past few years. And, not surprisingly, my answer has changed over that time. The skills required to develop and implement a commercial AI solution are very different today than they were a few years ago, and they will continue to evolve.

Changing Skill Requirements

Until recently, most engineers working in Artificial Intelligence had deep academic backgrounds and years of experience, often honed in a research setting. Their work required an advanced understanding of mathematics (including probability, linear algebra, and calculus) and computer science, and were employed creating highly-custom AI models to solve specific problems. With this specialized background, the number of such experts in the job market was low, and competition for these people was fierce. Consequently, building a team of AI experts was an expensive undertaking, and retaining them was challenging. This put AI development out of the range of many companies.

Over time, we've seen changes in the level of deep mathematical expertise required for many applications of AI. AI frameworks and open-source AI models have emerged, and this has a transformative effect on how many AI engineers worked. Rather than developing their own models from the ground up, AI engineers could use frameworks such as TensorFlow or PyTorch. They still needed to know what model to select and how the model functioned, how to create the right “features” in their data, and how to iteratively test and tune the model. But this knowledge was far less specialized and rarified than the deep mathematical expertise that was needed previously.

In the past several years, we've seen many AI workbenches, both open source and commercial, become available. These workbenches further abstract away some technical and mathematical details required to build a model. They allow software engineers and technically proficient business analysts to experiment with building AI-enabled solutions. With the recent wide release of Large Language Models (LLMs) such as ChatGPT, Llama, and Gemini, AI is becoming even more accessible to those who wish to deploy it in a commercial setting.

Using an LLM or another Generative AI tool, such as Midjourney or DALL-E, requires little technical or mathematical skill. Getting the results you want from these tools does require some knowledge, but it's a different skill set from that of more traditional forms of AI. There are limits to what a non-technical user can achieve with these tools. For example, if one needs to fine-tune an LLM or use an approach such as RAG (retrieval-augmented generation), more sophisticated technical skills are required.

For many “basic” AI use cases, advanced mathematical skills and AI expertise are no longer required. To be clear, there are and will remain plenty of situations in which the use of a packaged open-source model is not desirable or sufficient. In these “edge” cases, deep AI and mathematical experience will still be needed.

Parallels with the Software
Development World

We can look to the field of software application development to see how the skills needed for AI solution development might evolve. When I joined the field a few decades ago, there were many people like myself who had Computer Science degrees. There were coders who had other backgrounds, but many of the biggest software engineering tasks were handled by those with formal training in Computer Science. Today, by contrast, commercial Stack Overflow Developer Survey 2022 application development does not require a Computer Science degree. A study by Microsoft in 2012 found that 60% of working software developers did not have a CS degree, and more recently, the Stack Overflow 2022 Developer Survey showed that, while over 70% of professional software developers who responded had at least a Bachelor's Degree, these degrees included many non-CS scientific fields as well as Bachelor of Arts degrees.

The percentage of working software developers who have CS degrees has decreased over the past decades. Working software developers today do, of course, require specialized skills, such as training in general systems architecture, software development concepts and methods, and competencies in specific programming languages, databases, and other disciplines. But for most commercial application development, a Computer Science degree is not a requirement.

Where a CS degree is still valuable for software development is in the edge cases, those situations that lie far outside the normal application requirements. For example, engineers working on high-speed trading systems or self-driving vehicles do require profound knowledge and many years of training, which often includes rigorous academic training. These hard engineering problems constitute a small portion of use cases, compared to mainstream commercial application development, where these advanced skills are less in-demand.

Implications for AI

I can see AI following a similar pattern to that of application development, in which we see a bifurcation of the skills required. For the bulk of commercial AI development, we will see continued “productization” of AI tools. This means the level of academic, scientific, and mathematical knowledge needed to use those tools will continue to get lower. Using AI to create business solutions will require expertise, but not of the same sort required to build “edge” case advanced AI solutions.

But there will always be “edge” cases that require something far beyond what once can get by using commercially available or open-source AI models. And there will always be research into developing new AI techniques and algorithms. Transformer models and attention mechanisms that revolutionized Large Language Models are immensely complex, and were the culmination of decades' of work by brilliant and highly trained researchers. So there will always be a demand for highly trained mathematicians, engineers and scientists, even while the majority of business applications are built by those with different skill sets.

And what kinds of skills will be required to build commercial AI solutions? There's no single skill that will meet this need. But below are some of the capabilities that will be needed. Many of these are similar to skills that exist and are in demand today, suggesting that many companies can invest in upskilling current team members to help them adapt to the AI era.

AI Engineers

The most obvious role in AI solution development is that of the AI Engineer, or AI developer. As noted above, the level of scientific and mathematical background required for this role (in a typical commercial setting) is coming down. But it is still a specialized role that requires specific knowledge. Today, an AI Engineer needs to understand the concepts behind the AI tools they'll be using, with some proficiency in machine learning, deep learning, and neural networks. They also need to be familiar with the tools and frameworks you'll use for building models, such as TensorFlow, PyTorch, and Kubernetes. This should be complemented by an understanding of data, how to work with it, and its limitations. As you'll see below, there are many other roles required to deploy an AI solution, so the AI Engineer must have the “soft skills” to be an effective team member as well.

Solution Architects

The ability to bridge business and technology has always been important, and it will continue to be so. People who can work at the intersection of technology and business go by many names, such as a business analyst, a solution architect, a business architect. Sometimes they don't have a title that reflects this role. But regardless of title, they share the ability to understand business challenges and opportunities, and then work with technologists to design and implement a solution. This role will be as essential for AI as it is for traditional application development.

Data Experts

Another critical skillset for developing AI solutions is to understand the data that is used to train and provide input to AI models. It's no secret that most companies have spent far more time and attention on developing the functionality of applications than they have on the data capture and data management aspects of those applications. So, at least for the foreseeable future, there will be significant work required to acquire and make usable the data that AI models need. Data experts who understand data engineering, data architecture, semantics, and data quality will play an essential role in creating AI solutions.

Visualization and User Experience

We also should remember that most AI models, outside the realm of generative AI at least, create outputs that are not immediately digestible by business users.

For example, a prediction model may express its output as a numerical value with a range between 0.0 and 1.0. In most cases, that would not be immediately meaningful to the intended user. We will need experts in visualizing data and translating numerical outputs of models into a meaningful user experience. These team members will understand how this signal that comes out of an AI model will be used, and can transform that into the appropriate output for a business user.

This output could be a metric indicator, a number, or a graph or metric in a visualized data environment. The output could also be an SMS alert, an email notification, or a voice notification. Integrating the output of AI into the business process will continue to be an essential skill.

Business Process Expertise

We will also need experts in business process redesign or business process engineering, who understand how a business process can be optimized to take advantage of the capabilities of AI. We've learned many years ago that it's not always the best approach to simply automate an existing process through technology. Sometimes we need to redesign or optimize that process to take advantage of the way technology can support it.

Training and Change Management

As long as technology in general, and AI in particular, continues to evolve at a rapid pace, we will always have a need for people skilled at helping organizations adapt to change and acquire the new skills and competencies that they need. For sophisticated tools like AI, training, education, change management and related skills will be essential, especially as the pace of innovation continues to run quickly.

Other Areas of Expertise

These are certainly not all the skills necessary to support the implementation of AI solutions, but it should give a sense of the variety of skills and backgrounds that are necessary to create an effective and sustainable use of artificial intelligence. There are many other crucial roles that need to participate in AI system development, including cybersecurity, data privacy, technology infrastructure, and more.

Where To Find These Skills

Many companies have been tempted to start by hiring high-end data scientists and AI researchers. For the reasons described above, this may be overkill for many companies. More importantly, these highly skilled resources will not be able to work effectively unless you surround them with the additional skills necessary to build and secure the environment, build data pipelines, translate their work into a form consumable by the business, and more.

And for many companies, competing on the open market for highly skilled AI experts may not be the best use of their resources. Some of these companies should instead focus on building these skills internally, finding highly capable team members and providing them with the training and learning opportunities to build AI solutions. This learning process can be accelerated and made safer by augmenting the team with some expert help from a third party, with an emphasis on building your own team's skills and confidence. As an advisor and consultant myself, I'm aware that this last bit of advice may seem self-serving. Used correctly, an external resource can help your team grow more quickly. This is a very different approach than hiring an outside firm to build a solution for you.

AI is becoming a necessary core competency for most companies, and it is well worth your time to build up the strengths of your existing team. This allows you to capitalize on their knowledge of your business and offer them exciting new growth opportunities.