The Future Is Now: AI Agents in Action for 2025

In partnership with

The addition of the PC to the workforce in the mid-1980s and later profoundly impacted our productivity in immeasurable ways, email and word processing replaced snail mail and typewriters.

Later, the smartphone was reported to increase users' productivity by as much as 34%.

However, these advances will seem minuscule when we look at the next significant technology advance — AI Agents.

Here are just a few of the latest headlines from the last few days.

AI agents coming soon to a workplace near you 
AI Agents In 2025: What Enterprise Leaders Need To Know
AI agents may soon surpass people as primary application users

What does that mean to you?

Whether you are a manager or an individual contributor, you will likely manage teams of agents who can accomplish tasks on your behalf.

But if you are ahead of this trend, you can capitalize and take a formidable lead over your competition.

☕️ AI Tangle - Samsung's Potential Gemini Promo, Alibaba's AI Price Cuts & US Sanctions on Election Disinformation Groups 

🎯 The AI Marketing Advantage - Why Companies Are Investing Heavily in Purpose-Built AI

💡 AI CIO - Dear CIO: Welcome to 2025

The Artificially Intelligent Network (The AIE) is a collection of newsletters for business people looking to leverage AI.

2025 will be a transformative year, marking the rise of AI agents. In the long term, these autonomous entities promise to redefine industries, reshape workflows, and challenge entrenched norms. In the short term, they will be new entry-level additions to the workforce who will require the same training that their human counterparts might require, albeit in a slightly different way.

• One-third of consumers would rather purchase a product automated/digitally (e.g., through AI agents) vs. with a person.

• 39% of consumers are already comfortable with AI agents scheduling appointments.

• Over a third (34%) of consumers would work with an AI agent instead of a person to avoid repeating themselves.

Source: Salesforce

What are AI Agents?

AI agents are autonomous systems designed to achieve specific goals by perceiving their environment, processing information, and taking action—often without continuous human intervention.

Unlike traditional software, which follows static programming, AI agents adapt dynamically to new information and changing circumstances. Their autonomy and focus on specific objectives set them apart from other AI technologies, such as chatbots or virtual assistants, which are generally limited to predefined scenarios.

For example, while chatbots excel at structured interactions, such as answering customer inquiries or handling simple tasks, AI agents go further. They can integrate data from various sources, analyze patterns, and make decisions aligned with broader goals.

However, their reasoning ability remains limited, so tasks must be clearly defined and broken into manageable components. For instance, an AI agent tasked with optimizing a supply chain could analyze logistics data, anticipate disruptions, and recommend actions—but only within the boundaries of its specialized focus.

This limitation mirrors the structure of human organizations, where teams specialize in specific functions to achieve complex objectives collectively. Similarly, effectively deploying AI agents often involves creating "teams" of agents, each dedicated to one particular task, such as sales forecasting, customer support, or inventory management.

By combining their efforts, these specialized agents can address broader business challenges while ensuring that their individual roles remain aligned with their capabilities and objectives.

Training AI Agents: Just As Much Work as Training Humans

Deploying agents isn’t as simple as flipping a switch. Much like onboarding a new assistant, agents require training. This involves transferring institutional knowledge from humans to agents—a process that mirrors traditional hiring and onboarding. Individual contributors will become managers, orchestrating a team of specialized agents to achieve complex goals.

Managers, too, will evolve—overseeing orchestration agents that coordinate these digital teams. This shift redefines roles across organizations, embedding management skills into nearly every position.

Transforming Workplaces

AI agents are more than tools; they are collaborators. In 2025, they will handle routine HR queries, IT troubleshooting, and other administrative tasks, freeing employees to focus on strategic initiatives. However, questions about control and oversight persist. Striking the right balance between autonomy and supervision is crucial, especially in sensitive areas like finance and healthcare.

Trust in agents remains a key challenge. While agents are becoming faster and more reliable, many are not yet ready to grant them complete control—especially over financial transactions.

Navigating the Ethical Minefield

With increased autonomy comes heightened ethical concerns. Bias, transparency, and privacy challenges loom large. For instance, an HR agent optimizing hiring might inadvertently reinforce existing biases without careful design and monitoring. Worse, malicious actors could exploit agents for cybercrime or weaponize them for nefarious purposes. Robust governance frameworks are not optional; they are essential.

The Tech Behind the Magic

Advancements in cognitive architectures and specialized tools underpin this revolution, which makes agents more reliable and efficient. Multi-agent systems—where several agents collaborate—are becoming increasingly practical. Imagine a software development scenario where one agent writes code, another test it and a third debug it. This division of labor mirrors human project teams and unlocks new productivity levels.

Agents are also becoming multimodal, integrating video, audio, and text processing. Google’s Project Astra exemplifies this capability by processing video input to provide real-time assistance. Such technologies advance enterprise applications and empower individuals with disabilities, broadening access and inclusivity.

Challenges and Opportunities

Building effective agents is no small feat. Developers need new tools, frameworks, and benchmarks to refine performance and reliability. Training agents to perform complex tasks requires thoughtful design and robust testing.

We need to train these agents to be fully helpful. It’s not a simple task; just like humans, you must provide sound guidance to ensure your digital workers can handle the tasks you want them to achieve.

In addition, you have to be careful about giving them too much autonomy. Automation mistakes can create huge problems.

Take this example of automation gone disastrously wrong at Knight Capital Group. On August 1, 2012, Knight Capital caused a significant stock market disruption, leading to a large trading loss for the company. A misconfigured algorithm caused a massive stock market disruption, resulting in a $440 million trading loss in less than an hour.

Agents who are given too much autonomy and a lack of guidelines can become cautionary tales, so we will need to make sure that our

Looking Beyond 2025

The agent revolution is just beginning. While 2025 will lay the groundwork, the fundamental transformation will unfold in subsequent years. By 2026, hybrid human-agent teams will become commonplace, and companies that embrace this shift will outperform their peers.

• SmythOS - Smythe is a low-code agent builder that can start the creation of agents from natural language.

• Taskade - Taskade integrates AI capabilities into a collaborative workspace. It allows users to build, train, and deploy customizable AI agents with templates for specific roles.

• MindStudio—MindStudio offers an accessible, no-code approach to building intelligent agents.

• Langchain - LangChain is a prominent framework known for its modular architecture, advanced agent framework, and extensive tool integration capabilities. It is the most common framework for developers. 

• CrewAI - CrewAI is an open-source framework for developing and orchestrating collaborative AI agents. It features role-based agent design and structured workflows for coordinating tasks between multiple agents.

Prompts, Custom Instructions, and Knowledge to Build Your Agents

Agents today function as entry-level task workers. Combining agents into teams, they can function as an organizational unit.

While progressing toward greater autonomy, the most effective agents still require clear, incremental directions. They are highly specialized task workers focused on doing one thing exceptionally well. To maximize their potential, workflows must be broken into tightly defined tasks. Tools like RelevanceAI, MindStudio, and Taskade offer great starting points for crafting and deploying these agents.

Personally, I am using Taskade because of the interface. It allows collaboration across the agents and sharing with teams, but it’s by no means the only choice out there.

Key Guidelines for Building Effective Agents

1. Define Narrow Objectives: Specify exactly what you need the agent to accomplish. Broad prompts often lead to unfocused results.

2. Incremental Instructions: Break larger workflows into manageable, sequential steps.

3. Test and Iterate: Run small experiments with your agent’s outputs and refine instructions as needed.

4. Context is Key: Provide the agent background information and a clear understanding of the output format, including knowledge.

Build Your Own Agent

Most agents will use custom instructions and a series of prompts to do discrete tasks. So, take a complex task and break it down into logical steps. Create an agent for each step, including the orchestration of those agents.

They will probably start with a set of custom instructions. Think of this as your workplace code of conduct and standard operating procedures.

For example, we refer to our customers formally as Mr. or Mrs. Smith. We provide refunds only with valid proof of identity, and our preferred way of referring to our company is XYZ Company, the AI people.

Look at last week’s AI Lesson, Use ChatGPT Projects to Organize Your AI Workflows, to help create the organization that will make building your agents easier.

Prompt for The AI Agent

Then, each discrete task is given a prompt, just like we provide with ChatGPT. Here’s an example of how that might look for my researcher agent.

Each discrete task for an agent is a prompt, just like you use with ChatGPT. Here’s a prompt for an agent that scrapes the page and provides the output. Notice I am using the same prompting format with markdown and an example or two (few-shot prompting) to help guide the output.

### Prompt for the Agent:

You are a competitive analysis assistant. Your task is to generate a brief and focused summary of the web page provided. Include the following:

- **Name of Organization/Entity**
- **Primary Purpose or Mission**
- **Key Offerings (Products/Services)**
- **Unique Value Propositions**
- **Recent Updates or News**

Keep the summary under 200 words and emphasize relevance to competitive insights.

---

**Example Output:**

- **Name:** RelevanceAI  
- **Primary Purpose:** To simplify data analysis using AI-driven insights.  
- **Key Offerings:** Tools for embedding management, clustering algorithms, and automation for data workflows.  
- **Unique Value Propositions:** Focused on no-code solutions for data scientists and business analysts, with seamless integrations into existing analytics stacks.  
- **Recent Updates:** Announced a partnership with a major CRM platform, enabling direct embedding of insights into sales workflows.

Creating agents is more detailed than I can cover in this newsletter, but I hope you now have the pointers to get started. I will work on creating a class for those interested in learning more.

I appreciate your support.