As enterprises rush to adopt the technology, they are tasking their developers with building, customising, testing and deploying generative AI applications.

Yet enterprises underestimate the complexity of the AI stack and development lifecycle.

A new survey conducted by Morning Consult on behalf of IBM explores the complexity of developing generative AI applications, unveiling the challenges developers face when it comes to skills variance, toolset complexity, and accuracy assurance.

Skills Gap and Tool Sprawl

The survey, of over 1,000 enterprise AI developers in the US, revealed that generative AI skill levels vary significantly among surveyed developers.

A majority of developers who identify as ‘AI developers’ or ‘data scientists’ view themselves as experts in generative AI – but a minority of the seven other developer demographics do.

App developers in particular rarely view themselves as generative AI experts, despite being on the front lines of generative AI adoption.

Less than a quarter (24%) of application developers ranked themselves as ‘experts’ in generative AI.

This speaks to the skills gap in the generative AI space. For many developers, genAI is new terrain with a steep learning curve – and fast innovation means new technology is constant.

Compounding the skills gap is a lack of clarity when it comes to reliable frameworks and toolkits. Survey respondents listed the lack of a standardised AI development process as a top challenge, along with prioritising transparency and traceability.

“Lack of a standardized AI development process” and “Developing an ethical and trusted AI lifecycle that ensures transparency and traceability of data” are tied as top challenges in the development of generative AI applications among those surveyed.

Developers are also frustrated with the tools at their disposal. The most important tool qualities for building enterprise AI are also the rarest, respondents said, hampering the development process.

Meanwhile, developers must juggle a roster of tools.

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. Yet over a third of those surveyed also said those very same traits are the rarest.

A majority (72%) of those surveyed use between five and 15 tools to create an AI enterprise application. A notable number — 13% — use 15 or more tools.

Enterprises are investing in generative AI for a competitive advantage. An overly complex AI stack saps this investment and ripples out to other systems.

Recommended reading

• 72% of Companies Face AI Skills Shortage
• Only 10% of IT Leaders Have a Data Stack Ready for GenAI
• Demand for AI Skills Set to Skyrocket in 2024

These challenges will only become exacerbated as the industry pushes further into agentic AI, which promises greater power and autonomy – but also hinges on trust and integration with broader IT systems.

Almost all developers surveyed (99%) are exploring or developing AI agents – and the top concern reported for agentic development is trustworthiness.

A Solution? Simplifying the Stack

The survey sheds some light on what can be done to address the complexity of AI development.

While developers crave easy to use tools, only one third are willing to invest more than two hours in learning a new AI development tools. Simplicity and user experience is key when it comes to introducing new tools to the development process.

Further, 99% are using coding assistance in some capacity for Ai development, most commonly saving developers one to two hours a day, with 22% saying it saves them three hours or more.

Simplifying the AI stack and development lifestyle can be key in ushering the new era of AI safely and effectively.