Building AI with AI: Developers scramble to keep pace with new applications as security concerns grow – News

The emergence of generative artificial intelligence as a transformative force is reshaping how software is developed and deployed, and naturally developers are at the forefront of this transformation as they increasingly rely on AI for code building and the creation of AI agents.

Demand for AI agents, intelligent pieces of software that can perform specific tasks, is beginning to have a significant impact on the work of developers, and it is advancing AI capabilities more rapidly than many envisioned.

“AI assistants are doing so much more beyond code generation,” said Anoop Deoras, director of AI/ML for Amazon Q at Amazon Web Services Inc. “We have seen a few examples where AI assistants are outshining human capabilities.”

Moving agents into platforms and applications

Deoras spoke during a panel discussion on Wednesday at DeveloperWeek 2025 in Santa Clara, California, an annual gathering of enterprise developers and industry executives. AI and its many implications were a central part of the agenda for this year’s event, as developers sought to assess the promise and risk associated with a technology that’s moving at light speed through many organizations and sectors of society.

The growth of new applications for agentic AI has led companies such as DigitalOcean Holdings Inc. to retool its strategy as a competitor to cloud computing giants such as AWS and Microsoft Corp.

In January, DigitalOcean launched its new GenAI Platform designed to enable the building and deployment of AI agents in a matter of minutes. It is part of what the company believes will be a key battleground for cloud infrastructure providers, according to Bratin Saha (pictured), DigitalOcean’s chief product and technology officer.

“Today a lot of the action is in infrastructure,” Saha said during a presentation at the conference this week. “But we can see the action moving from the infrastructure layer onto the platforms and applications with agents.”

Examples of how agents are already making an impact on enterprise applications are not hard to find. What may be surprising to some is that AI agents are being rapidly adopted in technology risk-averse areas such as financial services.

Independent Bank Corp. is using a platform provided by SnapLogic Inc. to build AI agents for use cases such as fraud detection and help desk inquiries. “They were able to inject emotions for training their help desk personnel,” said Manish Rai, vice president of product marketing at SnapLogic. “These included emotions such as ‘act angry’ or ‘act annoyed.’ By using assistants, they were able to move a help desk worker from answering Level One questions to Level Three questions six to 10 times faster.”

Democratizing AI toolsets

Enterprise generative AI is also having an impact on the toolsets used by developers to build new applications. Key figures in the technology world are moving quickly to introduce capabilities for developers to build AI agents and integrate major AI models. What remains to be seen is how many of these tools will be proprietary versus open source.

An example of this can be found in Goose, an open-source AI framework released last month by Block Inc., the company led by Twitter co-founder and former CEO Jack Dorsey. As a freely available open-source tool, Goose is built to simplify the creation of AI agents and it supports the integration of leading AI models such as OpenAI, DeepSeek and Anthropic.

By releasing Goose as an open-source tool, Dorsey is seeking to democratize AI development and shape the evolution of AI agents using a community-based model.

“We think open source is what gets us to the best AI agent,” Jackie Brosamer, vice president of data and AI platform engineering at Block, said during a conference presentation on Thursday. “Ultimately the best version of Goose will come not just from the small team we have working on it, but from the community at large.”

Security concerns over LLMs

As key players maneuver for leading roles in agentic AI, another sector of the technology world is becoming increasingly concerned about the risks associated with building agents from widely distributed large language models. Earlier this month, a research team from ReversingLabs discovered a novel technique for distributing malware through files hosted on the leading AI model platform Hugging Face Inc.

That followed previous research that has documented breaching of model repositories in various AI as-a-service providers. One leading executive in the cybersecurity industry expressed concern that a wholesale shift to AI is not being accompanied by requisite caution.

“With LLMs, people assume that because it’s auto-generated code, it must be secure,” Checkmarx Ltd. CEO Sandeep Johri told News. “The assumption that it is secure code is a wrong assumption.”

Checkmarx, which provides scanning solutions for application security, correlates and prioritizes security signals in the enterprise development environment. Johri has seen examples where hackers flipped just one letter or number in a filename used for AI generation and then inserted malware that was subsequently downloaded by unsuspecting developers.

“Developers are the route to your codebase,” Johri said. “[Hackers] are targeting developers’ behaviors. I am a bit perplexed as to why more CISOs are not concerned about malicious code and scanning for malicious code. Doing a comprehensive scanning of the code is critical.”

The rapidly moving AI field, coupled with growing evidence of security threats, has underscored a continued need for human involvement. AI agents can already do a significant amount, but developers also realize that human supervision will still be a key requirement.

“These AI systems are going to become companions for software developers,” said Amazon’s Deoras. “How can you make sure the agent is behaving according to specifications? Having humans in the loop is extremely important.”

Photos: Mark Albertson/ News