July 30, 2025

The New Reality: How AI Agents Are Transforming Business Operations Today

Contents

AI agents are no longer a prospect of the future, with a far-fetched potential for value delivery. Today’s variants are moving far beyond low-level process automation; they can think, act, and coordinate goals independently, navigate complex enterprise environments, problems, and workflows, communicate with and control external tools, agents, and systems, and learn, adapt, and improve alongside their human counterparts. This isn’t hype anymore, it’s a reality, and one that we must understand how to navigate if we’re to pursue agentic transformation initiatives successfully and responsibly.

In this post, we’ll begin by investigating a series of agentic AI business trends that have emerged in the first half of 2025. Then, we’ll pivot, comprehensively discussing several business domains in which AI agents are currently providing value. We’ll conclude with some pragmatic insights and recommendations, focusing on what businesses should do to lay the groundwork and prepare for agentic AI integration and value delivery.

For readers who are unfamiliar with AI agents, we recommend checking out our five-part introductory series, in which we examine the following:

AI Agents: 2025 Business Trends

The AI agent revolution isn’t around the corner — it’s knocking on our door, with intensity, risk, and opportunity. While realizing how to best extract and operationalize AI agents’ value will not be easy, it will be necessary for businesses that aim to capitalize on this incoming wave of AI innovation.

To enable agentic innovation potential, however, businesses must understand what this technology is capable of, and most importantly, what sectors and domains it’s already transforming. We might be in the early stages of widespread agentic AI integration, but even with numerous challenges looming on the horizon and with much more to be discovered, we can still learn from what we see today:

According to a recent KPMG report, 65% of companies are currently piloting agentic AI deployments, with a startling 99% planning to integrate these technologies soon. While only 11% have done so already, it’s clear that agentic AI adoption trends are rapidly accelerating across industries. Deloitte’s findings further support agentic AI deployment trends, highlighting that 25% of enterprises currently using GenAI will deploy AI agents in pilot or production environments in 2025, with this figure rising to 50% by 2027. As for what comes next, it’s rather obvious: agentic AI integration challenges, and a little further down the line, failures.
KPMG also divides AI agents into four key categories, which should help enterprises comprehend their use case potential and application domain scope. These categories include:
- Taskers: Agents that are assigned a goal and can achieve it via a single or multi-task process. These agents are considered low-complexity; they possess rudimentary planning capabilities and are best suited for relatively simple task automation.
- Automators: Agents that can achieve and automate goals across a variety of interconnected systems and processes. They’re characterized as low-to-medium complexity agents with basic planning capabilities (e.g., can divide goal-based processes into sub-goals). Their value lies in their ability to reconstruct and redefine existing workflows and business processes while aligning goals across systems.
- Collaborators: Agents designed to work with humans to solve or achieve challenging problems and goals. They’re classified as medium-to-high complexity and showcase more sophisticated planning capabilities, defined by adaptiveness, context-awareness, and the ability to unite their efforts with those of humans. These agents are expected to be most valuable in domains motivated by creativity and innovation.
- Orchestrators: Agents that can cooperate and interact with multiple other agents across technology ecosystems to break down and reach multi-step, complex goals, effectively forming a multi-agent system. These agents are labeled as high-complexity and are the most advanced planners, possessing the ability to develop and orchestrate contingency strategies, facilitate collective action across multiple agents, and bolster resource optimization processes. For these agents, balancing disruptive potential with transformation will become a core challenge.
McKinsey has highlighted the “generative AI (GenAI) paradox,” a phenomenon characterized by high GenAI adoption rates coupled with low value realization. McKinsey attributes this phenomenon to the overprioritization of horizontal vs. vertical use cases.
- Horizontal use cases, where a business deploys AI across functions, do inspire benefits, but seeing as they’re distributed, measuring value delivery concretely remains quite difficult. However, horizontal use cases are significantly easier to pursue, requiring minimal change management and workflow transformation, with solid off-the-shelf solution availability.
- Vertical use cases, where AI is deployed within a specific business domain, despite being far less common (fewer than 10% move beyond pilot testing), can deliver direct and measurable economic gains. Nonetheless, they’re hindered by a variety of challenges, including siloed AI teams, immature pre-packed solutions, poor enterprise coordination, and cultural resistance, to name a few.
McKinsey is optimistic that AI agents will unlock the ability to scale and transform vertical GenAI use cases. They justify this expectation according to agents’ collaborative goal orientation and multi-task execution potential, adaptability, personalization, and elasticity benefits, and ability to redefine operational resilience through continuous monitoring, agile change management, efficient escalation protocols, and process automation. McKinsey also claims that agents will not only enhance existing revenue opportunities but introduce entirely novel revenue streams.
In Gartner’s view, AI agents will become embedded in one-third of enterprise software applications by 2028. Similarly, BCG predicts a compound annual growth rate (CAGR) of 45% for AI agents, such that by 2030, agentic AI will play a role in over 50% of tech applications. BCG further highlights four early-stage, real-world case studies that demonstrate cross-industry AI agent potential:
- Research & Development: By employing AI agents for lead generation, a biopharma company was able to cut cycle times by one-quarter while also reducing clinical report drafting times by 35%.
- Marketing: By leveraging agents to automate blog creation, one company diminished costs by 95% while enhancing publishing frequency by a factor of 50.
- Customer Service: A major bank that integrated AI agents for customer-facing interactions lowered customer communication costs by a factor of 10.
- Data & Tech: An IT department successfully transformed legacy technology infrastructures via agentic AI integration, elevating productivity by 40%.
Databricks CEO, Ali Ghodsi, stresses that fully automating complex, real-world tasks with AI agents is significantly more challenging than most would expect, adding that agents become exponentially more error-prone as workflows increase in complexity. This should bring some comfort to those who fear automation-induced job displacement, particularly since Ghodsi articulates a future in which human oversight and accountability remain critical while AI functions as an augmentative force. Hebbia’s (a VC-backed custom AI agent builder start-up) founder, George Sivulka, echoes Ghodsi’s sentiment, predicting that, “You’ll have hybrid teams of AI agents and humans,” envisioning collaboration over replacement.
Salesforce predicts that by 2027, agentic AI adoption will increase by 327%, inspiring an additional 30% increase in employee productivity, and driving HR departments to redeploy nearly a quarter of their workforce. However, this same study revealed that approximately 73% of workers don’t adequately understand how digital labor will impact their daily work, highlighting a massive internal communication and training gap.

So, given these trends and statistics, what are the bottom lines?

Enterprise agentic AI adoption is moving forward aggressively. However, mature deployment initiatives are lagging while integration challenges loom on the horizon.
To understand where AI agents can provide value, businesses should consider differentiating between their types and functions to ensure application domain relevance and effectiveness.
Due to their dynamism, adaptability, and complex task execution potential, AI agents could become a pivotal force for overcoming the scaling-motivated value realization challenges that vertical GenAI use cases inspire.
AI agents are already driving positive productivity and efficiency impacts across business sectors. Nonetheless, early-stage successful case studies don’t guarantee that agentic AI solutions will always unfold and function as intended.
Current business sentiments appear to favor human-AI collaboration and augmentation over automation. This trend should be interpreted with caution and skepticism; we don’t yet have enough data to make this assertion confidently, and these sentiments also exist in direct conflict with automation incentives.
The short-term success of agentic AI initiatives will require robust, agile, and cross-functional communication and training strategies that account for AI-induced workforce shifts and evolving business needs while preserving human agency and autonomy.

AI Agents: Realizing Business Value

Below, we’ll examine a selection of business domains and sectors in which agentic AI is already generating value, illustrated by early-stage, real-world case studies and use cases.

1. Customer Service & Support

Customer service and support processes are defined by a mix of high-volume, repetitive interactions, clear performance metrics, and significant labor costs — together, these factors have made this business domain a well-suited environment for showcasing quantifiable agentic AI ROI. In this case, AI agents are poised to fundamentally reshape the economics and strategic function of the modern-day contact center.

Case Studies

In April, Reuters reported that Verizon, after integrating an agentic AI customer service assistant, built on Google’s Gemini, experienced a whopping 40% increase in overall sales, attributed to major decreases in customer service reps' call times, which enabled them to spend more time on sales-facing tasks. This initiative effectively turned a traditional cost center into a powerful revenue-generation channel, and now, Verizon’s sales reps can comprehensively answer 95% of queries.
Eye-oo, an eyewear e-commerce platform, integrated an AI agent called Lyro to serve as its first line of customer support. Lyro reduced waiting times to just 30 seconds (from an average of 5 minutes), successfully handling roughly 80% of support conversations. The agent also drove significant sales benefits, acquiring over one thousand new leads and inspiring a 25% increase in overall sales, coupled with a 5x boost in conversions, generating an additional €177,000 in revenue.
Zolando, an EU-based online fashion retailer, launched a genAI-powered fashion assistant designed to ease customers’ product catalog navigation process while also supporting personalized outfit recommendations. This service-oriented agent directly enhanced pre-purchase customer engagement, facilitating a 23% increase in product clicks and a 40% growth in items added to customer wishlists.

2. Marketing & Sales

Marketing and sales are experiencing a radical, AI agent-driven transformation, with these steadily advancing systems automating and optimizing multiple components of the entire lifecycle, from content creation and brand awareness to lead conversion and sales execution. Over the last year alone, reputable firms have already begun reporting substantial ROI in the form of accelerated sales cycles, higher lead volumes, and dramatically improved campaign performance.

Case Studies

In 2024, JPMorgan Chase partnered with Persado (an AI vendor) to optimize its digital advertising copy, utilizing AI to generate and test different versions of ad copy. While it’s almost unbelievable, the bank managed to achieve a remarkable 450% increase in ad click-through rates, demonstrating how effective AI agents can be at balancing creativity with performance in content creation contexts.
Coca-Cola’s iconic “Share a Coke” campaign represents an early example of scalable AI-powered personalization, with the company having leveraged AI for mass data analysis across social media and sales channels to identify popular names for personalizing bottles. This strategy ultimately led to a 2% increase in sales and a colossal 870% boost in social media engagement, highlighting how agentic AI can streamline data insight generation for creative execution purposes.
Salesforce’s Einstein AI tools suite is quickly gaining notoriety. These tools can analyze millions of successful sales interactions, detect buyer behavioral patterns, automate A/B testing lifecycles, and identify deal stagnation signals, supporting many other functions as well. Overall, Einstein AI tools have been shown to cut average sales cycles by 37%, enable 47% faster deal progression for professional services firms, and achieve a 44% higher cross-sell success rate for financial services clients.

3. Software Development & IT Operations

While customer-facing agentic AI applications tend to capture the public eye more readily, AI agents are creating waves internally, redefining enterprise technology functions and operations. In software development and IT operations contexts, agentic AI is inspiring profound paradigm shifts, altering how enterprise technology is created, deployed, and managed, with internal transformation efforts yielding what can be interpreted as a “productivity dividend” in speed, quality, and efficiency, while simultaneously augmenting technical workforce experience.

Case Studies

In an enterprise study on the impact of AI coding assistants, Accenture discovered that its developers completed coding tasks up to 55% faster with GitHub Copilot while also noting elevated confidence in their work. The study further revealed an approximately 8% increase in pull requests (a measure of work output) along with an 84% increase in successful builds, indicating that developers were shipping more code, and that their code was actually improving — the impact on developer experience was profound, with 90% of developers reporting they felt more fulfilled in their jobs when using the tool.
By implementing AI-driven DevOps practices, SuperAGI achieved a 30% reduction in manual labor, a 50% increase in deployment speed, a 40% decrease in system downtime through predictive maintenance, and a 15% decrease in compute costs through intelligent resource allocation.

4. Finance & Banking

The financial services sector, which, historically, represents an early adopter of technology, has evolved into a generative AI champion, with 60% of banking and insurance executives reporting active use, and a remaining 38% indicating deployment interest within the next two years. High-interest use cases currently include fraud detection and personalized offers, and the AI fraud detection market alone is projected to surge from $7.5 billion in 2024 to $35 billion by 2032.

Case Studies & Use Cases

Employee & Advisor Productivity: To augment their human expertise and teams, major financial institutions are deploying internal AI agents. For example, in 2023, Morgan Stanley deployed a GPT-4-powered chatbot for its financial advisory teams, which was readily adopted by 98% of teams, demonstrably improving the quality and speed of client consultations.
Fraud Detection & Risk Management: AI agents are evolving into effective tools for dynamically identifying anomalies and mitigating risk. In fact, AI-enabled fraud detection systems can diminish false positive rates by up to 50%, allowing human investigators to focus their time and efforts on the most critical threats.
JPMorgan Chase deployed its “LLM Suite” to approximately 50,000 employees in its asset management division to boost research productivity and idea generation.
In early 2024, PayPal partnered with Feedzai (an AI firm) to integrate advanced fraud detection into its payment systems. However, beyond fraud, agents are further being developed to analyze complex regulations and corporate documents in real-time to ensure ongoing compliance, addressing a key concern for most, if not all, financial institutions pioneering AI initiatives.

5. Cybersecurity

Today, cybersecurity teams are facing a critical talent shortage, seeing a global shortfall of four million professionals, which could eclipse 80 million by 2030. Fortunately, AI agents now serve as an essential tool for augmenting overburdened security teams while countering increasingly sophisticated and proliferous AI-powered attacks, at both local and enterprise scales. By 2027, the AI cybersecurity market is forecast to reach 42.28 billion.

Use Cases

Analyst Workload Reduction: Agentic cybersecurity systems can autonomously detect and respond to attacks, investigate security alerts and anomalies, and generate incident reports, diminishing human security team workloads by up to 90%, effectively allowing human analysts to prioritize strategic threat hunting and response over triage.
Accelerated Threat Detection: AI is essential for accelerating efforts to identify and contain breaches; on average, it takes 292 days to identify a breach resulting from stolen credentials, which is a dangerously extended timeframe. Agentic AI systems that analyze network traffic and user behavior in real-time can dramatically reduce this detection window, drastically improving threat detection response times. In fact, on a daily basis, Google leverages AI to block over 100 million phishing emails.

6. Human Resources

HR departments are employing AI agents to streamline and automate a diverse conglomeration (e.g., payroll processing) of burdensome administrative tasks, effectively allowing HR professionals to concentrate on more strategic initiatives like talent sourcing, organizational culture transformation, and targeted upskilling.

Case Studies & Use Cases

End-to-End Process Automation: Agentic AI can augment and/or automate a diverse range of end-to-end processes across the employee lifecycle, from sourcing and screening candidates in recruitment to managing benefits administration and facilitating offboarding processes.
Internal Support Efficiency: In one quarter, IBM’s internal HR assistant, built on watsonx Orchestrate, saved a single department 12,000 hours by automating responses and processes related to common employee inquiries while also cutting processing times by a factor of two (from 10 to 5 weeks).
Quantifiable Productivity Gains: Several major corporations have reported substantial returns from HR automation. For example, Dell automated 30 HR processes, resulting in an 85% increase in overall HR productivity. Similarly, aerospace and defense company BAE Systems automated its payroll processing, bolstering data upload speeds by a factor of seven and saving over 2,600 hours of manual work annually. Furthermore, Coca-Cola automated its HR audit function, resulting in an added 16 extra hours of productive time to every workday and audit coverage that expanded to 100%.

7. Operations & Supply Chain

Throughout the complex and multi-faceted world of physical operations and logistics, AI agents are being utilized to generate advanced simulations (e.g., operational or supply chain disruptions) and optimize workflows, perpetuating major efficiency gains while enhancing supply chain resilience.

Case Studies & Use Cases

Digital Twins and Simulation: The BMW Group currently utilizes AI to scan its factories and create detailed 3D digital twins. These virtual models allow the automotive tycoon to run thousands of complex simulations, designed to optimize logistics and supply chain efficiency before real-world changes are implemented. In essence, digital twin creation allows companies to envision what certain solutions might look like when implemented before any physical efforts are undertaken, ensuring that interventions made are both pragmatic and effective. UPS is also in the process of building a digital twin of its entire global distribution network to improve package tracking and network planning.
Efficiency and Cost Savings: Toyota implemented an AI platform that empowered its factory workers to develop and deploy their own machine learning models, which led to a reduction of over 10,000 man-hours yearly. Moreover, Toyota’s mobility-focused subsidiary, Woven, achieved 50% total-cost-of-ownership savings in its AI-driven development of autonomous driving systems.

Conclusion and Recommendations

There’s no simple or direct playbook for how to extract and maintain agentic AI’s value in enterprise contexts. However, this doesn’t signify that enterprises can’t make certain strategic choices that will increase the likelihood that their AI agent initiatives remain effective, profitable, and relevant on both short and long-term timescales. In this respect, we provide readers with the following strategic recommendations:

Start by identifying the problem, not the solution. This may seem obvious, but it’s a mistake that many businesses continue to make; solution-oriented mindsets can undermine pragmatic, problem-solving and identification efforts while overlooking potential agent-induced risks, placing excessive priority on innovation without crafting an understanding of precisely what needs to be transformed to yield positive and measurable real-world impacts. When you start by identifying the problem, you also implicitly define or reveal your success criteria, which can then be operationalized as genuine metrics.
Don’t adopt AI agents for the sake of innovation. Building upon our previous point, agentic AI can be extremely alluring, especially when competitors pursue their own integration initiatives and manage to demonstrate tangible results that enhance their competitive edge. However, this shouldn’t be interpreted as a sign that you can’t match the competition by exploring and implementing less complex and costly AI solutions — adopt AI agents as a last resort, and in doing so, ensure that you have the necessary infrastructure, talent, and comprehension in place to support your initiatives.
Don’t copy your competitors because they “succeed”. The most effective AI solutions are those that operate within a highly concrete and targeted scope (don’t confuse this with a “narrow” scope), defined not by what works for others, but specifically, what works for you. Even if your business objectives align with those of your competitors, you may face a distinctly different selection of challenges, have advantaged or disadvantaged technological capabilities, or more or less AI resistance among key teams and personnel. Draw inspiration from your competitors, but don’t mimic them.
Map your workflow challenges independent of AI, and then remap them with AI in mind. Engaging in this process will allow you to identify where practical agentic AI solutions exist, categorized in terms of actual (not inferred) workflow pain points, while also envisioning what agent-specific workflow challenges (e.g., oversight gaps, reskilling requirements, security loopholes) could arise as a consequence of integration. Mapping AI-independent and dependent challenges will further provide you with a comparative reference point to return to when considering how to expand or fine-tune next-generation agentic AI initiatives.
Rigorously pilot test all AI agent initiatives. Agentic AI solutions may appear promising during early-stage deployments, however, AI performance can degrade significantly over time, elevating and accelerating compliance, security, and ethics risks before they can be appropriately identified, mitigated, and resolved. Pilot testing should commence with small teams tasked not only with continuously assessing the solution’s efficacy in real-world operational environments but also providing regular feedback to IT, security, risk management, governance, ethics, and HR teams.
Establish a third-party vendor risk management policy. If you’re deploying off-the-shelf AI agents (i.e., not building systems internally), you must vet your vendors and diversify your AI asset portfolio. On the one hand, you must ensure that your third-party systems are aligned with your business values, compliance requirements, and long-term strategic objectives, while on the other, you must recognize that full dependency on a single vendor could eventually compromise, if not destroy, your AI-enabled operations.
Stress-test your agentic AI solutions to identify critical security gaps. Formal red teams (specialized security teams tasked with probing a model’s defenses via adversarial strategies in controlled environments), internal or external, are essential to uncovering and patching security vulnerabilities, particularly as advanced AI systems undergo updates, modifications, and operational or environmental shifts. Red teaming is a continual practice; while it should play a foundational role in pilot-testing efforts, it must be vigorously pursued throughout an agentic AI system’s lifecycle to account for evolving adversarial threats and low-visibility, high-impact model risks (e.g., unintentional credential leakage).
Create a fallback plan for catastrophic AI agent failures. The most brilliantly orchestrated agentic AI initiatives can catastrophically and unexpectedly fail; even if failure probability is low, you must be prepared to maintain operations in the event of a catastrophic failure. A strong fallback plan should include measures for anticipating and counteracting agent-driven risks that cascade across interconnected systems or departments, strategies for dynamically resuming operations without incurring excessive overhead costs, mechanisms for deploying production-ready back-up systems (either as temporary or permanent replacements), and reliable methods for decommissioning and documenting compromised systems.

There are many more strategic recommendations we could offer, but seeing as this is the first in another multi-part series on AI agents, we’ll leave readers with one final input: don’t let your excitement about AI agents usurp your skepticism, and remember to approach all agentic AI initiatives from a deeply pragmatic, critical, and organizationally-specific perspective. Focus on what you need to do to succeed, and don’t let what others have done cloud your judgment.

For readers who enjoy our content, we recommend following Lumenova’s blog, where we cover numerous complex topics across a variety of fields, including governance, safety, ethics, innovation, literacy, and risk management. By contrast, for readers interested in building a more comprehensive understanding of the adversarial AI threat landscape and emerging AI capabilities, we suggest engaging with our AI experiments, where we run regular jailbreaks and capabilities tests.

For those who crave practical and scalable AI governance and risk management solutions, we invite you to check out Lumenova’s Responsible AI platform and book a product demo today. If this piques your interest, you might also want to take a look at Lumenova’s AI policy analyzer and risk advisor.

Make your AI ethical, transparent, and compliant - with Lumenova AI

Book your demo