Written by Ritika Gunnar | General Manager, Data and AI, IBM
January 8, 2025
IBM watsonx.ai helps simplify your AI stack.
New developer research details challenges developers face when building generative AI applications for enterprises, and suggests solutions
Developers across industries are seeing their job descriptions expand to include the popular phrase “generative AI.” As companies rush to adopt this technology, developers are responsible for building, customizing, testing, and deploying generative AI applications.
However, companies underestimate the complexity of their AI stack and development lifecycle. Beneath every sophisticated and intuitive AI application lies a complex and unwieldy technology stack.
Now, a new study sponsored by IBM and conducted by Morning Consult explores that complexity and explores the differences in skills, vast and complex toolsets, and the challenges developers face when it comes to ensuring accurate and reliable results from these systems. clarified the challenges faced by In our research, we interviewed more than 1,000 enterprise AI developers in the United States who are building generative AI applications for enterprises. Study participants occupy a variety of roles, including application developers, software engineers, and data scientists.
This study paints a clear picture. Enterprise AI developers are building tools to simplify the workflows of their colleagues. However, the development of these tools is often anything but. This research also provides clues to technologies and approaches that can help address this problem.
Challenge: Skills gap and tool sprawl
Our research reveals that generative AI skill levels vary widely among the developers we surveyed. The majority of developers who call themselves “AI developers” or “data scientists” consider themselves experts in generative AI, but only 7 of the other 7 developer groups consider themselves to be experts in generative AI. There are a few. App developers in particular rarely consider themselves generative AI experts, even though they are at the forefront of generative AI adoption.
Less than a quarter (24%) of application developers surveyed rated themselves as generative AI “experts.”
This speaks to the skills gap in the generative AI space. For many developers, this is new territory with a steep learning curve, and fast innovation cycles mean new technology continues to exist. The skills gap is further exacerbated by a lack of clarity around reliable frameworks and toolkits. Survey respondents prioritized transparency and traceability and cited the lack of a standardized AI development process as their biggest challenge.
“Lack of standardized AI development processes” and “developing an ethical and trustworthy AI lifecycle that ensures data transparency and traceability” are the top challenges in developing AI applications among those surveyed (33%, multiple respondents).
Developers are also frustrated with the tools at their disposal. Respondents say the quality of the most important tools for building enterprise AI is also the most scarce, hindering the development process. Developers, on the other hand, need to master a variety of tools.
According to those surveyed, the four most important qualities for enterprise AI development tools are performance (42%), flexibility (41%), ease of use (40%), and integration (36%). But more than a third of those surveyed also said those exact characteristics were the rarest.
The majority of those surveyed (72%) use between 5 and 15 tools to create AI enterprise applications. A notable number (13%) use 15 or more tools.
The results are clear. Developers are faced with the challenge of real-world complexity in AI stacks, and this has real consequences. Businesses are investing in generative AI to gain a competitive advantage. An overly complex AI stack wastes this investment and spills over to other systems.
These challenges will only get worse as the industry moves further into agent AI. Agent AI promises greater power and autonomy, but it also relies on trust and integration with broader IT systems.
Almost all developers surveyed (99%) are considering or developing an AI agent. The biggest concern reported regarding agent development is reliability.
It is clear that now is the time to address the challenges of AI complexity.
Solution: Simplify your stack (with the help of AI)
Our findings highlight what we can do to address the complexity of AI development, and some tools that are already helping. First, given the pace of change in the generative AI landscape, we know that developers are hungry for tools that are easy to use.
Only one-third of those surveyed were willing to invest more than two hours to learn a new AI development tool. This shows that simplicity and user experience are important when introducing new technologies to support the AI development process.
When it comes to developer productivity, the study found that the use of AI-powered coding tools has been widely adopted, resulting in significant time savings.
99% use some form of coding assistant for AI development. And most commonly, developers say these tools save them 1-2 hours per day (41% of developers), with some saying they save more than 3 hours ( 22% reported saving more than 3 hours).
Simplifying the AI stack and AI development lifecycle is a key focus for IBM. We know that with the right approach and tools, developers can leverage generative AI to drive business success.
We do this with big-picture strategies, including support for IBM’s open source AI. An open source AI stack means a more transparent, reliable, and innovative AI stack. We do something similar with certain products. IBM’s watsonx.ai, an end-to-end AI application development studio, offers a wide range of tools, frameworks, and integrations to help simplify and streamline your development lifecycle. IBM watsonx Code Assistant gives developers additional firepower when creating enterprise-ready AI applications. Additionally, IBM’s application integration solutions help developers create and manage APIs to modernize AI applications across hybrid environments. Meanwhile, the IBM Granite model provides an open source foundation for enterprise development of trusted AI.
The AI development stack doesn’t get much attention in the broader generative AI conversation. But it can play a huge role in the impact of technology. Make your AI stack as simple and intuitive as the applications it produces.
Developers looking for resources to support their generative AI projects should check out the new watsonx Developer Hub. It’s a central repository with the latest quick starts and guides for everything from models to chat to RAGs to tool calls and more.
