Remember HAL 9000 from "2001: A Space Odyssey"? The eerily calm, near-sentient robot? It was a dependable coworker for the humans on board. Well, at least before it turned rogue (more on that later).
HAL represents one of science fiction's most iconic visions of artificial intelligence (AI) with autonomy — an AI system that could observe, reason, and act independently to achieve its objectives.
That's a glimpse of agentic AI in action. Running on large language models (LLM), it works without human prompts. It not only creates and plans tasks but executes them on its own. Salesforce deems it "era-worthy," and McKinsey touts it as the "next frontier of generative AI."
"The AI agents market is expected to grow at 44.8% CAGR between 2024 and 2030, driven by technological advances in natural language processing (NLP)."
Markets and Markets
Backed by data from 3,621 reviews, this article explores how AI agents are being used in the workplace today and what that means for the future of work. By examining their most common applications, usage patterns, and other insights, we can understand not just where AI is making an impact but how it’s reshaping team structures, decision-making, and the division of labor between humans and digital coworkers.
Let’s get into it.
Agentic AI was first developed in the 2000s when machine learning (ML) models allowed agents to learn and improve using large databases. Today, the agentic AI landscape is based on advanced autonomy, operating with an ethical and responsible AI-controlled environment.
The key difference comes down to autonomy — the shift from reacting to commands to pursuing goals proactively.
Traditional chatbots and co-pilots are reactive tools that handle single tasks based on specific commands. Ask them to draft an email, and they draft it.
Agentic AI systems are proactive digital workers. Give them a high-level objective—"increase customer retention"—and they autonomously plan, execute, and adapt across multiple systems. Like Waymo, they navigate to the destination without constant intervention.
Largely, AI agents are autonomous digital workers that use tools to accomplish goals. These agents have the ability to remember across tasks and changing states, according to BCG.
But what does this look like in practice? Here’s an example: A human manager assigns an AI agent to create a blog post on digital marketing trends. The agent autonomously researches, requests graphics, drafts the content, stages it in the CMS, and schedules social promotion — all with minimal human oversight beyond final review. While this example demonstrates the autonomous capabilities of agentic AI, let’s understand how businesses use it today.
To understand the use of AI agents, we analyzed 3,621 reviews made by verified users worldwide on G2. We found that agentic AI is most commonly used in computer software, IT, and services industries, followed by financial services.
“AI agents are not just for the big established companies. According to G2 reviews from the past 12 months for AI agents, over half (55%) of reviews are from small businesses.”
Matthew Miller
Research principal at G2
Additionally, the fact that “AI chatbots”, “AI agents”, and “chatbots” are the most popular AI categories in terms of traffic on G2 testifies to the growing interest in agentic AI among users.
The emphasis on ease of use is a driving factor in agentic AI’s adoption. The democratization will accelerate its use across sectors and functions and reach non-technical users.
Users also value customer support, impressing the human-AI collaboration in onboarding and launching agentic AI solutions.
Stressing the decreasing time to ROI, Yukta Rustagi, a Market Research Analyst at G2, adds, “This also implies that organizations leveraging AI are gaining a competitive edge through efficiency and innovation. It reinforces the belief that AI agents are now a more immediate and impactful investment for businesses.”
ROI for AI Agents category (Jan 2023 - Dec 2024)
Source: G2 Market Research
Users have highlighted feature limitations in AI agents, but this reflects the technology's infancy rather than functionality gaps. We're still developing a shared understanding of what agentic systems should accomplish across domains.
While consumer solutions can rely on general features, enterprises need specialized agents that target specific use cases, such as coding, inventory management, and lead qualification. AI agents also present a unique customization challenge — they're adaptive systems that personalize through interaction, not configuration, functioning more like digital coworkers than traditional software.
Cost concerns signal low market maturity, as businesses struggle to quantify agentic AI value without standardized metrics or enough case studies to build compelling business cases.
While AI agents are being used across industries, our analysis of G2 reviews reveals that customer experience is the area where their impact is most pronounced. Nearly half of the user reviews mention CX-related improvements, making it a natural focus for understanding the real-world benefits of agentic AI.
While analyzing reviews for AI agents, we found that 217 of them, or 43%, mention customer experience (CX), making it arguably the most common area of impact for companies today.
Echoing the sentiment, Tim believes agents are the most employable in two use cases so far: customer support and sales development.
“Customer support teams currently face high backlogs, which agentic systems can quickly reduce. Sales development leaders see little downside to agentic SDRs, given their hard-to-fulfill quotas of lead generation, booked meetings, and additions to the pipeline,” he explains.
“In the next 10 years, AI in CX will reduce the cost to serve by an order of magnitude, enabling brands to expand touchpoints with customers in a way that has never been possible.”
Jason Maynard
Chief technology officer of AMER and APAC at Zendesk
Currently, 30% of consumers would work with an AI agent for faster service. “We predict that this will increase as the systems become more reliable and as users and businesses develop more trust toward the systems,” Jason says.
These immediate, tangible wins in customer-facing roles offer a preview of something more fundamental: agentic AI isn't just changing what work gets done — it's transforming how we think about work itself.
Agentic AI isn’t just about mere automation but cognitive reallocation. It’s creating a new way to look at the division of labor where humans are elevated to higher-order thinking roles.
To understand this real-world impact, we turn to industry experts who have observed AI agents in action, offering valuable insights into how businesses are integrating them, where human oversight is still crucial, and what skills will be needed in this AI-driven future.
Here’s what they have to say:
Some peddlers of AI agents are branding them as digital employees, others as teammates, and others still as tools that stand behind users as opposed to between them, says Matthew.
To this, Tim adds, “They require less human-in-the-loop efforts than LLM chatbots. We should think of these agents as team members that never take time off, get distracted, or develop bad attitudes.”
Mark Purdy, Director of Beacon Thought Leadership, says AI agents also perform a variety of specialized functions. For example, agents can gather information from multiple internal databases and external knowledge sources, assessing and synthesizing the insights for business analysts, lawyers, scientists, or other knowledge workers.
“AI agents can understand different business problems and contexts, triggering actions and workflows that reduce the strain on human workers,” points out Mark. For example, AI agents can assess email traffic from customers or clients, automatically responding to queries or complaints. They can monitor and follow up on sales leads.
“AI agents can help human managers and leaders make better decisions by running different scenarios or simulations to show the outcomes of alternative courses of action.”
Mark Purdy
Director of Beacon Thought Leadership
When asked how organizations must divide tasks between agents and humans, Mark says the degree of human involvement will depend on many factors. These include the decision's importance, the degree of trust in the AI agent’s recommendations, the consequences of a mistake, and the human worker's experience and judgment.
Echoing this sentiment, Sreelesh Pillai, Co-CEO at Zepic, notes that the company’s AI agents operate independently, mimicking and amplifying human capabilities while allowing businesses to configure human involvement as needed. But as AI agents gain more autonomy and capability, talk around AI agents being a threat to human workers is rising. Let us understand that.
Well, not really.
Agentic AI can, in theory, function autonomously and take over entire processes and systems. But will this integration be at the expense of human workers?
Concerns about AI are both justified and misplaced, believes Kate O'Neill, Founder and Chief Tech Humanist at KO Insights. “The threat isn't that AI will replace humans wholesale — it's that we might fail to reimagine work in ways that leverage uniquely human capabilities alongside AI.”
“The future of work isn't a zero-sum game between humans and machines. It's about creating synergies that make both more capable, more productive, and ultimately, more human.”
Kate O'Neill
Founder and chief tech humanist at KO Insights
Kate calls upon AI agent vendors to design their tools explicitly as human amplifiers, not human replacements. That means building tools that enhance human judgment, creativity, and emotional intelligence — the very qualities that make us uniquely human.
As agentic AI evolves, both sellers and buyers must implement it in a human-centric way. Intrinsic motivation, a key driver for employees, can take a hit if agentic AI is allowed to take over tasks that have provided workers with a sense of mastery and purpose. It’s no more about skill adaptation but reconstructing a professional identity for workers as work gets divided between them and AI agents.
However, protecting workers’ identity and motivation is only part of the equation. The autonomy that makes agentic AI powerful also introduces risks that can't be ignored.
According to Kate, the most pressing ethical concerns around agentic AI go beyond surface-level automation issues to fundamental questions of trust and decision-making authority.
“The core ethical challenge around Agentic AI isn't about algorithms or automation — it's about power.”
Kate O'Neill
Founder and chief tech humanist of KO Insights
Who controls these decisions? How do we ensure customers retain meaningful agency?
“Every time an AI agent makes a choice, it's essentially making a small prediction about human behavior and preference. Get enough of these micro-decisions wrong, and we're not just failing at customer service — we're undermining human autonomy. The stakes are higher than most companies realize,” warns Kate.
The solution? We need unprecedented levels of transparency with agentic AI. “Customers need to understand not just that they're interacting with AI, but how and why these agents make specific decisions,” suggests Kate.
The increasing autonomy and interconnectedness of agents introduce new technical risks, particularly when multiple systems collaborate to complete a complex task.
As we progress along the agentic AI gradient, agents will work with each other.
According to Tim, there are a few risks to watch for when this happens: they often need to exchange credentials to actually perform tasks within a multi-step process. That could pose security risks, as not all agentic platforms have the same level of trustworthiness.
“Reasoning errors (think hallucinations) have exponential impact as they spread across agentic teams,” says Tim. “Think of how statements can get distorted as repeated across a chain of human beings.”
In response to these valid security and trust concerns, leading providers are developing specific architectural safeguards designed to manage autonomous operations responsibly.
Responding to these concerns, leading AI agent sellers Salesforce and Zendesk claim their solutions feature security plug-ins beyond those traditionally deployed for AI tools. They say humans still control the wheel, customer data is safe, and workplaces are metamorphosing into more connected and productive spaces.
“Context is the king for accurate, personalized AI outputs,” says Leandro. “Without real-world data about your business and your customers, agent responses are generalized or, worse, rely on hallucinations and guesswork. Data is essential, but so is its secure and ethical handling.”
He explains that they developed the Einstein Trust Layer at Salesforce, which secures and anonymizes data to prevent leaks. “Transparency is also built into Agentforce. Those with digital labor on their teams can easily review the reasoning behind agent outputs and define the scope of agent responsibilities in natural language,” he adds.
In Zendesk's case, the human-in-the-loop approach is integral to using agentic AI.
Jason explains the approach and says they have configurable thresholds that allow human agents or administrators to review and approve AI-generated content and suggested actions.
“Any high-risk action, like issuing refunds or making account changes, can be configured to always have a human operator review and confirm it,” he adds.
As for governance around AI agents, Kate says, "Stop waiting for perfect regulations — they won't come. Instead, build governance frameworks that put human outcomes first."
“Yes, document your processes. Yes, establish clear accountability. But the real work is creating systems that amplify human potential rather than just automate human tasks.”
“Your ethics board should look like your customer base, not your executive team. Bring in the skeptics, the philosophers, the social scientists — and most importantly, representatives from the communities your AI systems will affect."
Kate O'Neill
Founder and chief tech humanist at KO Insights
The key skill of the future isn't writing prompts or managing AI — it's the ability to collaborate with AI to solve increasingly complex challenges.
Marshall McLuhan, a Canadian communications theorist, was prescient in his observation when he said, “We shape our tools, and thereafter, our tools shape us.”
That's exactly what's happening with AI.
We've created these tools to enhance our capabilities, and now they're reshaping how we work, think, and solve problems, says Sreelesh. It’s altering how intelligence itself operates across organizations.
Having the most advanced AI won't guarantee success — what will set organizations apart will be their ability to balance human and artificial judgment prudently.
This requires developing a way to orchestrate human intelligence and emotions into agentic AI-driven decision-making processes. This will not only delight customers but also help enhance employee experience.
Agentic AI can reason and act independently, while traditional automation simply executes predefined commands.
The main benefits include dramatically decreased time to ROI, increased organizational velocity of outcomes, 24/7 availability from an "always-on teammate," and improved customer experience through automated support and faster service.
Data privacy, hallucination errors, and lack of transparency are some of the key challenges — mitigated by governance frameworks like Salesforce’s Trust Layer.
New highly skilled jobs will emerge, often centered around designing, governing, and optimizing the AI systems. Examples include automation strategists, AI systems designers, customer journey architects, and technical CX professionals who configure the knowledge bases and policies that support the AI agents.
No. It will augment roles, turning professionals into strategists, orchestrators, and supervisors of digital coworkers.
Learn more about AI Agents in our latest report.
With inputs from Yukta Rustagi, Matthew Miller, and Brett Nehls of G2.
Did you know G2 has its own AI SDR with an average conversion rate of 30%? Try it out here.
Edited by Supanna Das
2024 was the year of tech giants touting the enterprise-readiness of AI software.
by Sidharth Yadav
At the Paris AI Action Summit in February, cracks around AI governance surfaced for the first...
by Sidharth Yadav
Generative AI is here to stay, and employees are using the new tech big time.
by Soundarya Jayaraman
