The hype cycle around artificial intelligence (AI) often moves faster than the capabilities it touts. But in 2025, we are witnessing a critical shift in the conversation: from predictive text generators to agentic AI¹—systems not just capable of reasoning, but of doing. The shift from passive to proactive intelligence is subtle in concept but profound in implication.

As investors, technologists and strategists, we need to recalibrate our understanding of what these AI agents are, what they can reasonably do and when we might expect returns on the billions of dollars pouring into this space.

What Are AI Agents, Really?

Definitions are still squishy, but a useful working model is this: AI agents are autonomous systems that can make decisions and take actions toward a goal. Not just chatbots that reply with flair, but entities that can reason through multi-step problems and act without constant human supervision.²

To borrow a metaphor from Capital One's chief AI scientist, talking about agentic AI today is like the parable of the blind men and the elephant—everyone is touching a different part.³ Some define it narrowly (assistive bots), others more broadly (fully autonomous digital workers). But the most agreed-upon litmus test comes from Gartner: does the AI make a decision, and does it take an action?⁴

Early enterprise deployments offer proof of concept:

• Adecco filled 100% of job openings for one client using agent-powered recruitment tools.⁵
• Visa, via ServiceNow, is running AI-powered dispute resolution systems.⁶
• The NHL is training agents to spot storylines in real-time games, like a player approaching a record-breaking moment.⁷

These aren't just productivity tools; they are beginning to behave like team members.

What Can They Do Today? And What Can't They Do?

Today's agentic AI excels in structured environments:

• Parsing resumes, matching candidates to job specs.
• Automating customer service resolutions.
• Surfacing statistical insights in real time.

Looking into Manus—a general-purpose agent from China—showed impressive results: it autonomously searched for journalists, parsed housing listings and curated nomination lists. It even admitted when it got "lazy."⁸

But its limitations are revealing⁹:

• It struggled with long tasks and unstructured prompts.
• It crashed frequently and needed constant clarification.
• Accessing paywalled content or real-time judgment calls are still out of reach.

In short: agents can handle projects you might give an intern, but not yet a chief of staff.

There's also a compute gap¹⁰. As agents move toward more complex reasoning, Nvidia estimates next-gen AI may need 100 times the compute resources compared to last year's models.¹¹ That's not a marginal hurdle; it's a generational infrastructure challenge.

What Should Investors Expect? A Timeline of Realism

In the near term—2025 to 2026—investors should expect agentic AI to find its strongest foothold in structured task automation. These are environments like HR, customer service and operational workflows where clean data and repeatable decisions are the norm. The most interesting plays here may be infrastructure and early enterprise deployment layers: companies like Nvidia (for compute), Salesforce (for enterprise orchestration) and vertical software-as-a-service (SaaS) platforms that embed agents directly into user workflows. Vertical refers to industry vertical and example companies could look like Veeva Systems (Health Care), Toast (Restaurants) and Procore (Construction).

Casey's General Stores, the third-largest convenience store chain and fifth-largest pizza chain in the U.S., has strategically integrated AI to enhance its pizza ordering and delivery services. In September 2023, Casey's implemented a conversational AI voice ordering system, known as the Automated Voice Assistant (AVA), across its entire network of over 2,500 stores spanning 16 states.¹²

Key Features of AVA:

• Order Processing: AVA efficiently handles incoming phone orders, accurately taking and confirming pizza requests without human intervention.
• Upselling Capabilities: The system proactively suggests additional items, enhancing the average order value.
• System Integration: AVA seamlessly connects with Casey's point-of-sale system and loyalty program, ensuring a cohesive customer experience.

The adoption of AVA was driven by Casey's commitment to reducing friction for both guests and team members by leveraging technology to simplify and streamline the ordering experience. It's useful to look at this example as an important company embarking on a journey to embed more and more technology-based solutions, even if it may not yet be directly using the most advanced AI agents today.

Between 2026 and 2028, we can reasonably expect agents to scale into semi-structured environments. This means more integration into business workflows that have some ambiguity but still operate within defined guardrails—think sales ops, supply chain optimizations and finance. During this phase, winners will begin to emerge among the orchestration layers—the platforms that can manage multiple agents across tasks and departments.

By 2028 to 2030 and beyond, we may start seeing autonomous agents reasoning across functional domains, coordinating decisions across departments, and even initiating cross-functional actions without human input. These systems could function like digital chief operating officers (COOs) or multi-domain analysts. The investment landscape by then may shift to full-stack agentic AI companies, but will likely also see consolidation as dominant platforms establish moats.

Investors betting on the "agentic turn" should avoid assuming a uniform disruption curve. Large language models (LLMs) are good at essay questions, not messy multitasking. Entry-level jobs that involve judgment calls, ambiguity, and human nuance are harder to automate than they look.¹³

But the shift is directional. Companies are already questioning whether to hire junior analysts or invest in agents. Apprenticeship models may re-emerge, where senior professionals train both people and machines.

AI agents are beginning to make notable contributions to business revenues and operational efficiencies across various industries. While the market is still evolving, several key developments highlight the growing financial impact of AI agents:

• Market Growth: The global AI agents market was valued at $5.4 billion in 2024 and is projected to reach $50.31 billion by 2030, reflecting a compound annual growth rate (CAGR) of 45.8%.¹⁴
• Enterprise Adoption: A significant majority of enterprises are integrating AI agents into their operations. Surveys indicate that 51% of organizations are exploring the use of AI agents, with another 37% already piloting them, underscoring their potential to enhance business processes and revenue streams.¹⁵
• Operational Efficiency: Companies implementing AI agents report substantial improvements in efficiency and cost reduction. For instance, AI agents have been shown to reduce customer service costs by up to 30% and reduce fraudulent activity by 40%.¹⁶
• Professional Services: Major firms in the professional services sector are investing heavily in AI agents to transform their business models. The Big Four accounting firms—Deloitte, EY, PwC and KPMG—are developing agentic AI platforms aimed at automating complex tasks such as financial management and tax compliance. These initiatives are expected to enhance productivity, reduce costs and potentially lead to new revenue models based on outcomes rather than billable hours.¹⁶

Conclusion: The Journey to AI Agents Is Just Starting

While these developments indicate a positive trajectory, it's important to note that the widespread financial impact of AI agents is still emerging. As adoption continues to grow and technologies mature, the business impacts are expected to become more pronounced in the coming years.

¹ Agentic AI is a type of artificial intelligence (AI) that can learn, make decisions, and act autonomously.

² Source: Isabelle Bousquette, "Everyone's Talking about AI Agents. Barely Anyone Knows What They Are," Wall Street Journal, 3/29/25.

³ Source: Bousquette, 3/29/25.

⁴ Source: Bousquette, 3/29/25.

⁵ Source: https://www.staffingindustry.com/news/global-daily-news/adecco-turning-to-salesforce-ai-agents-for-recruitment

⁶ Source: https://www.servicenow.com/company/media/press-room/visa-dispute-management.html

⁷ Source: https://www.greenfly.com/resources-category/customer-showcase/nhl-ai-power-digital-media-access/

⁸ Source: Caiwei Chen, "Everyone in AI Is Talking about Manus. We Put It to the Test," MIT Technology Review, 3/11/25.

⁹ Source: Chen, 3/11/25.

¹⁰ A "compute gap" refers to a disparity or difference in computing resources or capabilities, often related to factors like hardware, software, or access to computational infrastructure.

¹¹ Source: Daniel Howley, "Nvidia's Huang Isn't Shying Away from DeepSeek, Says AI Needs 100x More Computing Power," Yahoo! Finance, 3/19/25.

¹² Source: "Casey's General Stores to Rollout SYNQ3 Restaurant Solutions Conversational AI Voice Ordering,"news release, 9/6/23.

¹³ Source: John Burn-Murdoch, "Why Hasn't AI Taken Your Job Yet?" Financial Times, 3/28/25.

¹⁴ Source: Grand View Research Report Summary. https://www.grandviewresearch.com/industry-analysis/ai-agents-market-report

¹⁵ Source: KPMG AI Quarterly Pulse Survey: 2025 Is the Year of Agentic AI. This most recent survey was conducted 11/7/24–12/9/24.

¹⁶ Source: Wayne Butterfield, "AI Cuts Costs by 30%, but 75% of Customers Still Want Humans – Here's Why," ISG.

¹⁷ Source: https://www.businessinsider.com/deloitte-ey-launch-agentic-ai-platforms-big-four-competition2025-3