Artificial Intelligence (AI) agents have emerged as a transformative force in enterprise technology, promising enhanced efficiency and automation. However, the rapid proliferation of AI agents has led to a mix of auspicious developments and inflated claims. Tech headlines are declaring 2025 the “year of the AI agent,”[1] but it’s important for IT and business leaders alike to cut through the hype and understand what is, and what is not, realistically achievable with AI agents today. This piece builds on our recent white paper on automating workflows using LLMs and AI agents,[2] and it aims to provide business leaders with a clear understanding of AI agents, market trends and strategic considerations for investment over the next six to 12 months.

Understanding AI Agents

AI agents are autonomous software programs designed to perform specific tasks or make decisions based on data inputs. They leverage technologies such as large language models (LLMs) and other AI/ML models to interact with users, systems or even other agents to achieve defined goals. Agents can range from simple task-specific bots — for example, a customer support chatbot in Microsoft’s Copilot or IBM’s Watson Assistant — to complex multi-agent systems that collaborate to solve intricate problems, as seen in some of OpenAI’s research agents.[3]

How do agents differ from traditional generative AI (GenAI)? Models like OpenAI’s ChatGPT, Google’s Gemini or Anthropic’s Claude excel at GenAI tasks — e.g., producing text, images or code from a single prompt. AI agents go a step further by combining reasoning, memory and objective-based autonomy. Rather than purely generating content, AI agents are designed to take actions or perform a sequence of steps based on information in order to complete a broader objective. For instance, a GenAI model might draft a marketing email, while a chain of AI agents could draft the email, schedule its delivery via a CRM, and monitor performance, all with minimal or zero human intervention. This distinction is critical for business leaders: GenAI is a powerful tool for content creation, but AI agents act as an operational layer that can automate workflows and drive decision-making across enterprise functions.

Market Overview

Technology companies and startups alike have begun to capitalize on the pervasiveness of AI agents. According to IBM and Morning Consult, it is estimated that greater than 95 percent of developers are actively developing or experimenting with AI agents. Furthermore, Reuters reported that the cloud computing giant Amazon Web Services (AWS) has doubled down on its investment in LLM-derived AI agents, forming an entirely new business unit focused on AI agents, along with a slew of enhancements to its Bedrock and SageMaker services to help customers build and integrate agents in their work.[4] These announcements come alongside many other tech giants also racing to be competitive in the AI agent market.

With this backdrop, business leaders are rightly curious about the broader market outlook for AI agents. According to MarketsandMarkets, the global AI agent market is projected to grow from USD 5.1 billion in 2024 to USD 47.1 billion by 2030, at a CAGR of 44.8 percent. North America currently dominates the market, accounting for approximately 40 percent of the global share in 2024.[5] Critically, ROI is driving adoption: early enterprise deployments of AI agents have yielded up to 50 percent efficiency improvements in functions like customer service, sales and HR operations.[6] In other words, organizations investing in agent automation are already reporting significant cost reductions and productivity gains in key areas.

Major Players and Market Outlook

As the AI agent ecosystem rapidly matures, we see the market evolving into two primary categories: agent frameworks and agent providers. Frameworks offer the building blocks to develop custom agents, giving enterprises flexibility, control and extensibility. This includes open-source tools or libraries designed for developers and data teams to tailor autonomous workflows to specific business needs. We see hands-on developers and engineers as the primary users of agent frameworks.

On the other hand, agent providers offer turnkey AI agents-as-a-service, often built on proprietary LLMs and cloud infrastructure. These provider solutions focus on ease-of-use, vertical specialization and faster time-to-value, but they typically sacrifice some control and adaptability. Business leaders should assess both approaches based on factors like scalability, ecosystem maturity, security/compliance and alignment with internal capabilities.

Below is a comparison of building with a framework vs. buying from a provider:

Agent Framework vs. Agent Provider

To make informed decisions, leaders should evaluate their needs based on their organization's technical maturity, IT support and strategic goals. Frameworks offer flexibility and long-term control, but they require a larger initial investment in time and resources with a technical team to develop and tune the agents. Providers deliver speed and immediate impact, but do not allow the same adaptability as custom agents and tend to be more expensive over a longer period. In practice, many successful organizations will likely adopt a hybrid approach — piloting quick wins with provider solutions while simultaneously developing internal expertise via frameworks — to capture near-term value without compromising long-term differentiation.

Hype vs. Reality

There is significant, and often justified, promise that AI agents can transform how businesses operate and manage their workflows. However, AI agents are not always the best solution for a problem. It is important to consider the following challenges and limitations with AI agents:

• Reliability: LLMs are prone to hallucinations and other inconsistencies. Chaining multiple AI-driven steps in a single workflow can compound these issues if guardrails are not set to combat this — especially for tasks requiring exact outputs, for example, retrieving factual quantitative information or making high-stakes decisions.
• Performance and Costs: Very large and state-of-the-art LLMs, such as OpenAI’s GPT-4o, Amazon Nova Pro (and beyond), and Claude Sonnet, tend to have high degrees of accuracy on their own. However, when working with multiple agents or tool/function calling, they are very slow and expensive. This presents a problem when chaining agents together into a complex agent workflow, as often looping and retry logic is required, which can quickly escalate compute costs.
• Legal Concerns: Companies may be held liable for the mistakes of their AI tools, and agents are no exception. For example, consider the recent chatbot blunder in which Air Canada’s chatbot provided incorrect information regarding its bereavement fares. A court ultimately ruled that Air Canada was at fault, and it was ordered to pay back the customer.
• User Trust: This consideration is shared with a broader generative AI implementation, not just agents using LLMs. As we have seen, AI models, but especially generative AI, have a "black box" nature, in which the reasoning and decision-making is obfuscated from the user. AI agents add another layer of complexity and make it hard for users to understand and trust their outputs. It is critical to gain user trust, especially for sensitive tasks or tasks providing exact information. Techniques such as providing citations, producing transparent step-by-step explanations of what the agent is doing, and applying post-processing verification of answers can all help foster user trust.

So what should business leaders do to successfully and effectively integrate AI agents into enterprise operations? Organizations can start this journey by assessing their specific needs, identifying where agents can provide the most value, such as automating repetitive workflows, enhancing customer support or accelerating data-driven decision-making. Launching small-scale pilot programs allows teams to validate the technology’s effectiveness, refine use cases and build internal confidence before scaling up.

Equally important is conducting a rigorous vendor evaluation, focusing on factors like reliability, data security and compliance with industry standards. As AI agents become more autonomous, organizations must establish human oversight frameworks to ensure accountability and maintain trust.

Finally, a comprehensive cost-benefit analysis should guide deployment decisions, ensuring AI agent investments are strategically aligned and deliver meaningful ROI.

Tactically speaking, once use cases and a vendor or tool are selected, we recommend combining tightly constrained LLMs with good test data, keeping a human in the loop for oversight, and following traditional engineering best practices to achieve reliable results in automating complex tasks.

Where to Start

For most enterprises, we recommend starting with low- to medium-complexity use cases with tasks that are repetitive in nature but require human intervention for domain knowledge. This allows teams to learn, iterate and build trust while minimizing risk. As confidence and experience grow, more advanced and high complexity use cases can be incorporated with a clear understanding of where agents can provide efficiencies and where human oversight remains essential. Below is an example list of low- to medium-complexity use cases broken out by business function:

Customer Service and Support

1. Appointment Scheduling: LLM-powered agents can handle customer interactions to schedule appointments in real time — accessing schedules, confirming availability and managing changes with minimal human intervention.
2. Customer Support Ticket Management: Intelligent agents can streamline help desk ticket processing by classifying issues, prioritizing them based on urgency or customer value, and routing each ticket to the appropriate team members. With capabilities like semantic search through knowledge bases, historical issue resolution matching, and workload balancing, these agents can make support operations more responsive and data driven.

Sales and Marketing

1. Lead Qualification: AI agents continuously analyze incoming lead data to score and qualify leads. Using this data, they match leads to the most suitable sales representatives based on any defined rule, such as territory, product knowledge or workload. This can help accelerate the sales cycle and improve conversion.
2. Marketing Campaign Optimization: Agents monitor defined campaign performance metrics across channels, autonomously adjusting targeting, messaging and spend. With direct access to LLMs, these agents can also generate personalized content, enabling rapid A/B testing and real-time optimization.

IT and IT Support

1. IT Service Management: Manage incidents, problems and change requests end-to-end with agents by triaging tickets, suggesting resolutions based on past data, and coordinating workflows across systems. This can help to reduce resolution time and improve uptime, without additional strain on IT resources.
2. Cybersecurity Monitoring: AI agents proactively monitor networks for potential threats, using modules such as anomaly detection and pattern recognition to flag risks. Agents trigger automated alerts, enforce security protocols, and can even initiate mitigation actions to create a more resilient cyber defense posture.

Finance and Operations

1. Invoice Processing: Agents equipped with optimal character recognition (OCR), pattern recognition, and exception-handling logic process invoices from ingestion through approval. They flag anomalies, validate entries and update finance systems, drastically reducing manual workloads while ensuring compliance.
2. Expense Report Summarization: Agents analyze and summarize expense reports by extracting key line items, detecting outliers or policy violations, and generate concise summaries for Finance review. Since humans remain in the loop, this is a low-hanging use case that can help to accelerate reimbursement cycles and enforce spend governance, while retaining stakeholder oversight and trust.

Conclusion

At Alvarez & Marsal (A&M), we help organizations move beyond experimentation and into scalable impact with AI agents. Leveraging our library of pre-built assets, accelerators and solution templates, we fast-track agent deployment across key business functions. Whether it’s building custom solutions with flexible open-source frameworks or implementing enterprise-ready solutions through our strategic partnerships with vendors like AWS and Microsoft, our team provides the technical expertise and platform integrations needed for end-to-end implementation. From tailored agent design to orchestrated deployment within your existing tech stack, A&M’s AI and Analytics team empowers organizations to operationalize AI agents in a matter of weeks — accelerating value, not just timelines.

Ready to turn AI into action? Our team at A&M helps organizations build, deploy and scale AI agents across your enterprise. Connect with us to learn how we can support your AI initiatives.

Contact Us

Sources: