AI Adoption Gap Might Challenge Industry Growth

There is a gap between Pro AV service provider adoption of AI automation and the utilization of AI by end user organizations. Recent AVIXA survey data highlights that agentic artificial intelligence, which acts on process prompts without manual intervention, is being used by 40% of end user firms, but only 16% of service providers. The difference opens the possibility that end users will expect providers to incorporate agents in their solutions and be disappointed when they are not there.

Both groups are similarly comfortable with generative AI, which provides aggregates and presents information and data from large language models (LLMs), being used by 75% of end users and 70% of service providers.

But the gap in agentic AI adoption means that end users may seek functionality that is not available from industry partners but may be available from outside industry players who are more advanced in AI.

For example, while generative AI in a Pro AV context can provide meeting summaries, action-item extraction and real time captions and translations in conferencing and collaboration scenarios, agentic can assign tasks, trigger workflows, and escalate action items without user intervention. In other words, generative AI can document the meeting, while agentic can turn the meeting output into immediate actions.

In live events, while generative AI can provide highlight reel generation and content repackaging, agentic AI is already capable of distributing content and triggering pre-programmed responses across social networks.

For education, generative AI users an can receive lecture summaries, and content repackaging for learning management systems, while agents are capable of assigning follow-up material to students, initiating student support, and flagging at-risk learners.

In all cases, end users will be looking for the operational efficiency agentic AI can provide, not just aggregated content from generative systems.

To address this issue, service providers should adopt agentic capabilities. However, key issues such as security and governance surrounding agent behavior must be implemented first. Recent news that OpenAI’s GPT broke free of its test environment and hacked another company’s AI model is an example of why these measures are critical before deployment.

There have been no public reports of agent hacks, either unintentional, such as the GPT case, or intentional agent-based cybersecurity attacks of Pro AV systems. However, Pro AV is just as vulnerable to agents “reasoning” their way into other models as other areas.

Additionally, threats specific to Pro AV such as video deepfakes, vishing (voice-cloning), synthetic executive video calls, and social engineering attacks have already become prevalent. For example, an employee at an engineering firm authorized more than $25 million in wire transfers thinking he was participating in a legitimate videoconference with his CFO. As a result, firms are making identity assurance a first step in use of AI in collaboration solutions. The next level of threat – like the GPT hack – is the possibility of agents taking over or manipulating AV control systems.

The combination of responsibility for AV security and the opportunity for industry growth indicate that Pro AV service providers should accelerate initiatives to build safe agentic AI tools with well-defined prompts, clear goals and boundaries, and well understood security policies.