Building generative AI applications is too hard, developers say

When everyone touts the benefits of generative artificial intelligence, telling the world how much easier it will make all of our lives, many developers who actually have to build those miracle apps are of a different opinion, according to a new survey sponsored by IBM.

The survey gathered input from more than 1,000 enterprise developers in the US, who enumerated the challenges they face in creating generative AIapplications.

Process challenges

Unsurprisingly, one key issue is the skills gap. Less than one quarter of application developers (24%) consider themselves experts in generative AI when asked to rate their proficiency and professional experience in the field. Among the seven self-identified categories of developer surveyed (AI developer, data scientist, software engineer, system developer, machine learning (ML) engineer, software developer, IT engineer, AI engineer, and application developer), the only groups in which a majority considered themselves experts in generative AI were the AI developers and data scientists. Less than half of those styling themselves as ML engineers (43%) and AI engineers (38%) view themselves as generative AI experts.

“This speaks to the skills gap in the generative AI space,” said Ritika Gunnar, general manager, Data and AI at IBM, in a blog post. “For many developers, this is new terrain with a steep learning curve — and fast innovation cycles mean new technology is constant.”

But even those whose AI skills are up to scratch face challenges, Gunnar noted. “Compounding the skills gap is a lack of clarity when it comes to reliable frameworks and toolkits,” she said. “Survey respondents listed the lack of a standardized AI development process as a top challenge, along with prioritizing transparency and traceability.”

Each of those two categories was cited by one third of survey respondents, making them the top challenges among the 10 challenges reported. They were closely followed by customization to accommodate enterprise context (32%), rate of change of techniques/technology (31%), infrastructure complexity (29%), and establishing governance and compliance (28%).

Large language model (LLM) quality was considered a challenge by only 19% of respondents.

However, more than one quarter of developers (26%) faced every programmer’s nightmare: lack of clarity on business outcome/objective.

Inadequate tools

The majority of developers, the report said, use between five and 15 tools to do their jobs — 35% use five to 10, 37% use 10 to 15, and fully 13% use 15 tools or more. Yet those tools don’t always meet their needs.

“Performance (42%), flexibility (41%), ease of use (40%), and integration (36%) are the four most essential qualities in enterprise AI development tools, according to those surveyed,” Gunnar noted. “Yet over a third of those surveyed also said those very same traits are the rarest.”

Around one third of respondents also bemoaned the state of four other essential traits: documentation quality, cost-effectiveness, community support and resources, and that the tools be open source.

Given the number of tools they need to do their job, it’s no surprise that developers are loath to spend a lot of time adding another to their arsenal. Two thirds of them are only willing to invest two hours or less in learning a new AI development tool, with a further 22% allocating three to five hours, and only 11% giving more than five hours to the task. And on the whole, they don’t tend to explore new tools very often — only 21% said they check out new tools monthly, while 78% do so once every one to six months, and the remaining 2% rarely or never. The survey found that they tend to look at around six new tools each time.

While more than half of developers are leveraging low code (65%) and no code (59%) tools, pro code tools still rule, at 73%. Nearly all respondents are using AI coding assistants in their AI development work, with 41% saying it saves them one to two hours per day.

Agent worries

The final area the survey explored was agents. Pretty much everyone (99%) is exploring or developing AI agents, and the concerns are predictable: 31% worried about trustworthiness (ensuring outputs are accurate and bias-free), 23% were concerned about introducing new attack vectors that bad actors could leverage, 22% cited compliance and adherence to regulations, and 22% worried that agents could become overly autonomous and humans lose oversight and visibility into the systems.

Regardless, the top three use cases are predictable: customer service and support (50%), project management/personal assistant (47%), and content creation (46%). Only 1% of respondents were not exploring agentic use cases.

A call to arms

The survey highlights the fact that, while AI and generative AI are becoming increasingly important to businesses, the tools and techniques require to develop them are not keeping up.

“Our survey results shed light on what we can do to help address the complexity of AI development, as well as some tools that are already helping,” Gunnar noted. “First, given the pace of change in the generative AI landscape, we know that developers crave tools that are easy to master.”

And, she added, “when it comes to developer productivity, the survey found widespread adoption and significant time savings from the use of AI-powered coding tools.”

This means, she pointed out, that to build those increasingly important generative AI applications, the development stack needs a lot more love.

“The AI development stack doesn’t receive a lot of attention in the broader generative AI conversation. Yet it can play an outsized role in the technology’s impact,” she said. “Let’s make the AI stack as simple and intuitive as the applications it produces.”