DEEP DIVE

AI Agents Are the Next Big Thing

Today’s AI assistants, like ChatGPT, perform tasks through prompts, but they can’t act independently. This is about to change with AI agents—or agentic AI—capable of completing complex tasks autonomously.

Gartner predicts “Agentic AI” will be a top technology trend in 2025, and forecasts that by 2028, 33% of enterprise software will include agentic AI, up from less than 1% today. Companies like Salesforce, HubSpot, Google, and Microsoft have already launched tools for creating task-based AI agents, signaling the start of AI’s third wave. Nvidia and Visa envision futures where employees manage dozens of AI agents each.

Remember that WOW moment when you used ChatGPT for the first time? I think you are likely to go through that again with AI agents. 🤯

Let’s dive.

What Are AI Agents?

AI agents are software that can autonomously complete tasks by understanding context, planning workflows, and executing actions. Unlike AI assistants like ChatGPT, which assist users based on prompts and require continual input, AI agents can take the initiative, working through multi-step processes with minimal human intervention.

Imagine spending hours writing a brief, creating content, scheduling social media posts, analyzing data, and generating reports. AI assistants can help, but still require constant guidance. Now imagine delegating these entire workflows to AI agents that work autonomously. Huge difference, right?

AI agents represent a major leap forward in autonomy, but collaboration remains key - humans set rules, goals, and workflows while agents execute them. While current AI agents are task-specific, major developments are accelerating their capabilities: Anthropic has launched its "Computer Use" feature in beta, with OpenAI and Google expected to release similar tools by 2025, enabling AI to directly operate computers and handle multi-step workflows that previously required significant human oversight.

Just as humans are active participants in our environments, AI is beginning to embed itself into workflows and take meaningful actions. While current levels of AI agency remain relatively low, they are set to grow rapidly.

 The Spectrum of AI Agency

AI agency exists on a spectrum. On one end are systems that perform simple tasks under strict conditions. On the other are advanced agentic AI systems capable of reasoning, learning and making decisions completely autonomously. See real use cases below.

To help organizations benchmark and compare AI agent solutions, several industry players have developed their own scales to measure levels of agency (examples here, here, and here), drawing inspiration from the Society of Automotive Engineers (SAE) framework for assessing levels of driving automation.

When examining these scales, it becomes evident that today’s AI agents are far from achieving the higher levels of autonomy. Based on the framework below, most current agents operate at Level 2 or 3, indicating they are still in the early stages of development.

Agents vs. Automation

Unlike automation, which follows predetermined recipes for specific scenarios, AI agents can adapt to new contexts and solve problems in real time. Automation relies on rigid rules, whereas agents leverage reasoning to handle exceptions and achieve goals.

For example, automation might stumble if a process deviates from predefined steps. An AI agent, however, can pivot, find alternatives, and complete the task. This adaptability makes agents ideal for managing complex workflows, where flexibility and context are critical.

As agentic AI matures, its ability to autonomously manage entire workflows will make it easier to distinguish it from traditional automation. Andrew Ng, a recognized leader in AI, illustrated this difference sharing an example recently where an AI agent navigated unavailable online services by autonomously switching to alternatives to achieve its goal.

Why It Matters for Marketers

AI agents have the power to transform workflows by streamlining repetitive tasks such as data collection and report creation. This allows users to dedicate more time to higher-value, strategic activities that require human expertise. While AI assistants and chatbots introduced over the past two years have promoted similar benefits, true AI agents go far beyond, offering significantly greater advantages, such as:

1. Purposeful: Unlike freeform tools, agents follow specific workflows tied to a set of tools to complete an activity, measurable against existing goals and KPIs.

2. Accessible to Non-Developers: With no-code and low-code platforms like Microsoft Copilot, Salesforce Agentforce and RelevanceAI, non-technical users can create and manage agents.

3. Contextual: Specialization, predefined prompts and access to relevant data enable agents to deliver precise, actionable responses compared to traditional AI models.

For example, a marketing analyst tasked with creating a report on generative AI trends could delegate the process to a multiagent system. The interface agent (orchestrator) would define the scope, while specialized agents handle data collection, content summarization, and report formatting. This not only streamlines workflows but ensures accuracy and alignment with business goals.

Maximizing the ROI of AI Agents

The true power of agentic AI emerges when multiple agents work together, much like high-performing teams leverage diverse expertise to tackle business challenges. However, realizing this potential requires more than just implementing technology - it demands a strategic approach that considers people, processes, and technology in equal measure.

To prepare for this hybrid human + machine future, organizations should:

1. Build Employee Buy-in: Proactively discuss the transition to a hybrid workforce, grounded in responsible AI principles

2. Redesign Workflows: Map how AI agents will complement existing teams and processes

3. Ensure Seamless Integration: Verify compatibility with existing AI frameworks and tools to enable true autonomy

4. Enable Cross-System Connectivity: Deploy vendor-agnostic solutions that work across enterprise systems and departments

5. Audit Current Tools: Identify agentic AI opportunities within existing software and platforms

6. Leverage Data Intelligence: Leverage high-quality data to power deeper reasoning and decision-making

7. Start Small, Scale Smart: Begin with controlled experiments to understand capabilities and user acceptance

8. Establish Governance: Create oversight frameworks to track progress and manage deployment risks

The Takeaway

As excitement around agentic AI grows, it's essential to stay grounded. While the technology's potential is significant, success requires clear strategies, thoughtful integration, and continuous experimentation. Organizations and professionals who begin their AI journey now will be best positioned to thrive in this emerging landscape.

Despite speculation about AI replacing jobs, today's agents are simply tools for delegation - not autonomous replacements for human workers. Humans remain central to the process: defining rules, setting constraints, designing workflows, and establishing success criteria. Agents then execute within these carefully constructed boundaries. Given that most business processes still require human judgment or aren't even digitized yet, this human-directed approach ensures that people remain in control of how and where AI agents create value. This "human in the loop" model isn't just a safety measure - it's the key to unlocking AI's true potential.