While everyone praises generative AI for how much more convenient it makes our lives, there are many different opinions among developers who should create these ‘miracle apps’, which is why the IBM-sponsored new Survey resultsrevealed in
In this survey, we surveyed more than 1,000 corporate developers in the U.S. about the various challenges they face while developing generative AI apps.
Problems with the process
As expected, the skills gap is a major issue. In the question asking people to evaluate their proficiency and professional experience in the field of generative AI, less than 24% evaluated themselves as generative AI experts. When developers are divided into seven job groups (▲Generative AI developer ▲Data scientist ▲Software engineer ▲System developer ▲ML engineer ▲Software developer ▲IT engineer ▲AI engineer ▲Application developer), they are evaluated as specializing in generative AI. The only job categories were AI developers and data scientists. Even among ML engineers (43%) and AI engineers (38%), less than half consider themselves experts in generative AI.
Ritika Gunnar, IBM Data and AI General Manager blogMentioning this, “This clearly shows the technology gap in the field of generative AI. For many developers, this is new territory and requires a steep learning curve. “Furthermore, the rapid innovation cycle means that new technologies are constantly emerging.”
But Gunnar pointed out that even developers with sufficiently advanced AI skills often encounter difficulties. “One of the factors exacerbating the skills gap is the lack of reliable frameworks and toolkits. “Respondents cited the lack of a standardized AI development process as a key challenge, along with prioritizing transparency and traceability.”
These two items (lack of a standardized AI development process and prioritizing transparency and traceability) were mentioned by one-third of survey respondents and ranked top among the top 10 challenges reported. These are followed by adapting to business context (32%), pace of technological change (31%), infrastructure complexity (29%), and governance and compliance (28%).
Meanwhile, LLM quality was identified as a challenge by only 19% of respondents, and more than 26% of developers said they faced what is every programmer’s nightmare: ‘lack of clarity about business outcomes/goals.’
inappropriate tool
According to the report, most developers use between 5 and 15 tools to get their work done. 35% of respondents said they use 5 to 10 tools, 37% said they use 10 to 15 tools, and 13% said they use more than 15 tools. However, these tools do not always meet developers’ needs.
“The most important characteristics of enterprise AI development tools were performance (42%), flexibility (41%), ease of use (40%), and integrability (36%),” Gunnar said. “However, more than one-third of respondents said this trait was the rarest one.” About one-third of respondents also expressed dissatisfaction with four other essential characteristics: quality of documentation, cost-effectiveness, community support and resources, and whether the tool is open source.
Considering the number of tools required for development work, it’s not surprising that companies are reluctant to spend a lot of time adding new tools. Two-thirds of respondents (66%) said they were willing to spend no more than two hours learning a new AI development tool, 22% said they were willing to spend three to five hours, and only 11% said they were willing to spend more than five hours.
Additionally, developers overall do not explore new tools often. Only 21% of respondents said they check for new tools every month, 78% said they check for new tools once every 1 to 6 months, and the remaining 2% said they rarely or never check. It was found that when exploring a new tool, people look at an average of six things.
Meanwhile, more than half of respondents are using low-code (65%) and no-code (59%) tools, but Pro Code tools still dominate (73%). Almost all respondents are using AI coding assistants in their AI development work, with 41% saying it saves them 1 to 2 hours a day.
Agent Concerns
The final area of the survey was agents. Nearly all (99%) of respondents said they were exploring or developing AI agents, and concerns about this were predictable.
31% were concerned about reliability (ensuring that output is accurate and unbiased), and 23% were concerned that it would become a new attack vector that malicious actors could exploit. Additionally, 22% were concerned about regulatory compliance and compliance, and another 22% were concerned about agents becoming too autonomous, resulting in a loss of human oversight and visibility into the system.
Nonetheless, the three most popular use cases were as expected. Customer service and support (50%), project management/personal assistants (47%), and content creation (46%) are top examples. Only 1% of respondents said they were not exploring agent-related use cases.
call for change
The survey highlights that while AI and generative AI are becoming increasingly important to businesses, the tools and technologies needed to develop them are not keeping pace.
“The survey results suggest ways to help address the complexities of AI development, and are already introducing some helpful tools,” Gunnar said. Considering the rapid pace of change in the generative AI environment, it is clear that developers are craving tools that are easy to learn. “We also see widespread adoption of AI-based coding tools and significant time savings when it comes to developer productivity.”
This means that more attention and effort will be needed in the development stack to build increasingly important generative AI applications.
“In the broader discussion around generative AI, the AI development stack doesn’t get enough attention,” Gunnar said. But this stack can play a very important role in the technology’s impact. “We need to make the AI stack as simple and intuitive as the applications it creates,” he added.
dl-itworldkorea@foundryco.com
Source: www.itworld.co.kr
