Advancements in artificial intelligence (AI) and machine learning (ML) have been dominating the headlines for months. ChatGPT, Alphabet’s Bard, and other big tech AI/ML initiatives are all recurring players in emerging AI technology. While these newer AI/ML products indeed represent an upward shift in AI/ML’s role in the technology ecosystem, there’s also been a much less discussed, much more subtle shift happening on the other side of AI/ML. The tools, processes, and structures that develop and manage AI/ML are likewise seeing a rapid, often chaotic, evolution. These tools have major implications for the deployment of AI/ML, especially as a myriad of non-technology focused companies launch their own AI/ML solutions.

The Evolution of Spaghetti Code

There was a time when software was easier to develop and manage, with hundreds instead of millions of lines of code. Back in 1969, Apollo 11 was built on 145,000 lines of code, while today’s luxury car has more than 90 million. Open-source tools have aided development, but in many ways added to complexity. Open source started as a fringe activity but has since become the center of the universe of software development. Open source has progressed and expanded through a dramatic evolution from its creation with the GNU project in the early 1980s at MIT, to the launch of GitHub in the late 2000s, to the acquisition of Red Hat by IBM for $34 billion.

Yet, while software development has gotten more and more complex, there is a growing shortage of computer science graduates and related skillsets in the market. In fact, according to the U.S. Labor Department, there will be 1.2 million unfulfilled software engineering jobs by 2026, growing to 85.2 million globally by 2030.

The hiring cycle for developers can be long and candidates can be in short supply, but companies now have established best practices to guide developers, standard processes to ensure quality outputs, and tools to streamline certain processes. The convergence of these tools and standards is referred to as their Developer Experience, or DevEx for short. The importance of DevEx cannot be overstated — not only does it have direct implications for developers, but because those developers are creating software their companies rely on, it has direct implications for a company as a whole. Many companies are in dire need of improving their DevEx to keep pace with the new tools and open-source libraries their engineers have access to.

DevEx and Data Management for AI and ML

Today’s AI/ML applications are so complex, they’re often impossible to reverse engineer. Many companies want to manage AI/ML development the same way they manage traditional software development. Understandably so as they both rely on abstract, complex thinking and extremely expensive labor. But the two processes are distinct. While AI/ML does incorporate elements of software development, its roots in data make the core AI/ML work much more comparable to data science and statistical analyses, but with its own unique nuances.

Machine learning scientists must align their work with business goals, wrangle and clean data, develop and train models, manage the deployment of data models, and monitor said models’ real-world results. Each of these steps represents a unique challenge. Data wrangling for example, can take weeks of an ML scientist’s time as they go from data source to data source — an expensive challenge given the labor cost of ML scientists.

Monitoring models similarly poses a unique challenge. How can companies ensure their models don’t drift against business interests, or worse, against legal and compliance interests? Despite these challenges, companies still want to join the AI/ML wave. While the benefits of AI/ML may vary by use case, they tend to hold great promise, and no company wants to be left behind. Nonetheless, AI/ML DevEx, and AI/ML operations in general, require custom solutions.

As one may expect however, not every company is approaching this the same way and AI/ML DevEx varies widely. While the established big tech companies have years of experience managing AI/ML, central infrastructure teams to oversee AI/ML, and relatively streamlined processes (albeit, still evolving), most companies have a long way to go to reach a mature, evolved stage of DevEx.

More traditional companies have tried varying strategies. Some have outsourced their AI/ML development to specialists and consultancies at the expense of internal competency development. Others have leveraged third party tools that provide a set of AI/ML infrastructure, and as impressive as many of these tools may be and are becoming, dependency on these may not align well with long run business interests. Similarly, many are experimenting with different governance structures to see which are best able to facilitate AI/ML development while still aligning with business and legal compliance goals- often finding a truly streamlined and efficient process difficult to achieve.

While these companies may look to big tech as examples, they likely do not have the resources to mirror them, nor the expertise to get there. Thus, a different path may be required. A unique story is being played out in more traditional companies, whose AI/ML will have strong cumulative impact across industries and use cases. As such, how these companies navigate the changing AI/ML landscape and how AI/ML tool and services providers respond to their needs will ultimately dictate a larger piece of the potential value of AI/ML than most would expect. ◆

Stream our latest Future Perfect Tech podcast