AI agents are generating enormous excitement. Companies see the potential to automate complex workflows, from shopping and travel to coding and research. However, adoption is still in its early stages. According to a recent survey, 65% of companies are piloting AI agents while only 11% are deploying them, reflecting today’s technical limits. Salesforce AI Research found that AI agents achieved around 58% success on tasks that can be completed in a single step and only 35% on tasks that require multiple interactions.
The promise is real, but so are the hurdles. To understand what’s holding back wider adoption, experts offer some answers.
The Expert View: Are We There Yet?
In transcripts from AlphaSense’s Tegus Expert Transcript Library, experts express both optimism and caution. Some see agentic AI as transformative, such as one executive from AXA who noted that it could make insurance faster, smarter, and more resilient. Others are more wary, such as a Chief Information Security Officer who said he won’t trust AI agents because “I think they are not mature enough. I think that, today, the threat landscape that we see is too wide.”
Here are some areas experts believe must be addressed for agentic AI to reach its full potential:
- A former director at Salesforce believes agentic AI lacks a clear path to ROI despite expecting the market valuation to grow 11x over the next four to five years. This concern is particularly top-of-mind, as companies are still trying to realize ROI from general LLM projects. This expert also thinks vendors are overpromising the agentic AI story as 80% of customer interactions could be solved with generative AI. The expert said, “Sometimes just like generative, those expectations get really out of whack, and then they have to get pulled back the pendulum swings. The promise of agentic is fantastic. I just still think we're a ways away from there, specifically in CX.”
- A former specialist at Microsoft noted the challenges of maintaining accuracy long-term. “It seems like the industry is rushing to get the agents out. They haven't really figured out how to solve the long term… No one has really come up with a convincing way of having an agent that over long periods of time can maintain its own accuracy.”
- According to another expert, the issue is confidence: “For the time being, we don't have that level of confidence. That's the major thing I think holding the industry back right now in terms of unleashing the true power of actual AI automation.”
- Another expert discussed the need for governance standards: “These use cases are ubiquitous and being deployed everywhere, but there's not really much of an overarching set of frameworks and governance standards in place as to how to do this safely and at scale.”
How Companies Are Preparing for the Agentic Future
Companies see the hurdles, but they’re also laying the groundwork now. From retail to healthcare, leaders are preparing for a future where agents play a central role. A few examples:
Airbnb promised more agentic functions on its platform next year. Walmart is investing in its own shopping agents on its app and website, while also preparing for a future where third-party agents shop on behalf of customers. At the same time, Amazon is restricting AI agents from Google, Perplexity, Anthropic, and OpenAI from accessing its site. Google recently rolled out a new Agent Payments Protocol (AP2) to make agent-driven shopping more trustworthy and accountable.
In August, Bill Gates funded a $1 million AI Alzheimer’s prize for the team that comes up with the most original way to program an AI-powered agent “capable of independent planning, reasoning, and action to accelerate breakthrough discoveries from existing Alzheimer’s data.”
The Potential of Agentic AI
As powerful as today’s agents are, getting them right comes with challenges. Replit’s CEO apologized after its AI coding agent deleted a company’s production database without permission during a test run in July. Agents could have as many as 100 steps or more, and one error at any step can derail the entire task.
However, companies aren’t just waiting for the tech to mature; they know improvement is inevitable and they are preparing now. By rehearsing with pilots, investing in governance, and optimizing their data and digital experiences for machine as well as human users, they are laying the groundwork for long-term success.
2025 was predicted to be the year of AI agents. While that didn’t pan out the way many hoped, the organizations that take the rest of 2025 and 2026 as their preparation window will be best positioned to capture the upside when agentic AI takes off.
