The Hidden Challenge of AI Agents: Institutional Knowledge

Meta’s Mark Zuckerberg recently suggested that AI-driven systems could replace many mid-level coding roles at Meta.

This possibility underscores a broader trend: large language models from OpenAI, Anthropic, and Cursor excel at writing code because they draw on millions of open source examples—much as human developers do when learning from GitHub.

The paradox is that those who contributed code might have helped build the tools that now threaten their own positions.

Yet, not every role is vulnerable. Specific tasks lack the clear outputs machine-learning models need for training and rely on institutional knowledge no algorithm can replicate.

In these cases, we’re replacing tasks, not people. To that, these people will train agents and, for the foreseeable future, act as managers and orchestrators of AI employees.

This week, I’ll tackle some of those challenges and put you on the path to building your own robot army.

☕️ AI Tangle - Biden's Last AI Chip Regulations, Anthropic's Next $2 Billion & xAI's "Unhinged Mode" On The Horizon

🎯 The AI Marketing Advantage - A No-Fluff breakdown of AI Agents For Marketers

💡 AI CIO - GenAI Revolutionizes Cybersecurity

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

The Hidden Challenge of AI Agents: Institutional Knowledge

AI agents will offload your menial tasks, but they will require training just like human workers

AI Agents hold immense promise for businesses. They’re tireless, punctual, and don’t require benefits or vacations. Yet, there’s a hidden challenge that could undermine their potential: the transfer of institutional knowledge.

The market for AI agents is projected to grow from USD 5.1 billion in 2024 to USD 47.1 billion by 2030. This notable expansion, characterized by a Compound Annual Growth Rate (CAGR) of 44.8% between 2024 and 2030, is mainly driven by technological advances in natural language processing.

Institutional knowledge is what human employees carry with them over years of experience. It’s not written down. It’s not in the CRM. The unwritten rules, subtle preferences, and nuanced decisions make businesses thrive.

For instance, a virtual agent will not know that a key client always prefers a phone call to confirm major shipments over an email. It will not instinctively acknowledge a customer’s wedding or recall their preference for receiving holiday greetings. And it will not understand the intricate, undocumented product details that come with years of human expertise.

This is a major hurdle. While agents excel at pattern recognition and structured workflows, they lack the intuition and reasoning to navigate these softer skills. Simply mining data from a CRM system won’t bridge this gap.

Humans Training Agents

To make agents truly valuable, businesses must focus on training them beyond standard datasets. Humans must take on the role of trainers, imparting the subtleties of relationships, preferences, and contextual knowledge that agents can’t learn independently.

There’s another piece to this puzzle. Agents aren’t exceptional at reasoning—yet. They’re more like smart workflow systems than independent thinkers. To maximize their utility, tasks must be broken into discrete steps, with workflows meticulously designed for them to follow.

In this sense, training agents mirrors how we train human employees. It’s not just about inputting data but teaching processes, behaviors, and nuances. As agents gain more “agency” in the future, these skills will become even more critical.

Here’s the kicker — even individuals with no management experience will need to learn these training skills. Leading a team of AI agents will become as essential as managing human employees is today.

The market for AI agents is growing rapidly. Businesses that can transfer institutional knowledge effectively will gain a significant competitive edge. Those who won't find their agents falling short, unable to handle the real-world complexities of their operations.

How To Get Started Training AI Agents

Training employees to use AI agents begins with a clear understanding of fundamental concepts and an awareness of how those agents integrate with daily work. By building core AI skills, practicing precise prompting, and dissecting workflows into manageable tasks, teams learn which responsibilities are best suited for automated assistance.

Retaining high-impact tasks for human focus while delegating repetitive or data-heavy jobs to AI is the next step. Ongoing oversight ensures these agents remain accurate, consistent, and aligned with business objectives.

Build Core AI Skills

Employees must begin with a solid grounding in AI essentials. This includes understanding how models process data, recognizing the limits of AI outputs, and exploring ethical guidelines. Offer accessible training programs from reputable institutions or online platforms. Encourage hands-on learning with small projects—these experiences reinforce concepts and highlight practical use cases.

Practice Effective Prompting

Once employees grasp basic AI principles, focus on refining their prompt-writing techniques. Clear queries and instructions lead to more accurate agent responses. Teach employees to test different prompts, evaluate outcomes, and iterate until the agent delivers usable results.

Break Down Workflows

Encourage teams to dissect their everyday tasks into discrete steps. Identify which tasks are tedious or error-prone—these are strong candidates for delegation to AI agents. Retain high-value, strategic activities for humans. A methodical approach helps avoid confusion when distributing duties between employees and their digital assistants.

Delegate and Manage

Effective management of AI agents requires consistent oversight. Assign owners who monitor agent outputs and address discrepancies promptly. Set performance benchmarks and establish a feedback loop. By systematically reviewing outcomes, employees can refine inputs and keep agent tasks aligned with business goals.

The Key To Agentic Success

Promote a culture of continuous improvement. Encourage employees to share lessons learned, host internal demos, and discuss fresh approaches. Update training materials and hold regular refresher sessions to ensure alignment with new AI developments. And encourage everyone you know to subscribe to the AI newsletters and courses at TheAIE.net.  

Implementing these steps equips teams to train, oversee, and continually improve AI agents. By doing so, businesses gain a workforce adept at blending human insight with digital horsepower. The success of AI agents depends not just on their intelligence but on our ability to teach them what they can’t learn from data alone.

• Google’s Project Mariner - A research prototype exploring the future of human-agent interaction, starting with your browser.

• AI SDR - AiSDR is an AI-driven sales development platform designed to automate and enhance various sales tasks traditionally handled by human Sales Development Representatives (SDRs). It streamlines processes such as lead generation, personalized outreach, follow-ups, and lead qualification, enabling sales teams to operate more efficiently and at scale.

• CrewAI - Simplify Automation, Amplify Results with CrewAI The easiest way to deploy powerful multi-agent automation,built to handle your most complex use cases and deliver rapid results

Using ChatGPT as your Search Engine

OpenAI has enhanced ChatGPT with a feature called ChatGPT Search, which enables real-time web searches directly within the chat interface. Although I was initially unimpressed, I am now more enthusiastic about it. 

ChatGPT's search functionality integrates hyperlinks directly into its responses, allowing users to access original sources seamlessly. ChatGPT provides information sourced from the web, including inline citations with clickable links, enabling users to verify and explore the content further. Additionally, a "Sources" button at the end of responses compiles all referenced links for easy access. In the above image, you can see that

ChatGPT allows users to engage in natural, dialogue-based queries, enabling more intuitive information retrieval. This conversational approach facilitates follow-up questions and clarifications, enhancing the depth of understanding.

Accessing ChatGPT Search:

Initially rolled out to ChatGPT Plus and Team subscribers, OpenAI plans to extend access to all users over time. You can also replace your default search in Chrome using the ChatGPT Search Chrome plugin.

Enter your question or prompt as usual. ChatGPT will determine if a web search is necessary based on your query. Alternatively, you can manually activate the search by clicking the web search icon.

ChatGPT will provide answers with citations and links to relevant web sources, allowing you to verify and delve deeper into the information.

While ChatGPT aims to provide accurate and current information, it's advisable to cross-reference responses with the cited sources to ensure reliability. ChatGPT adheres to content guidelines and refrains from generating explicit or inappropriate material.

You can use natural language or a prompt for these searches. In this case, I asked ChatGPT to use sources I liked. Otherwise, you are probably fine just using natural language as you would with Google.

# Role 
You will act as a researcher for the latest productivity tips for marketing professionals. 

# Instructions 
Use reputable sources and search OpenAI.com, ProductHunt, Wired.com as well as Reddit.com and X.com for the latest tips and tricks to improve the productivity and quality of marketing for B2B Companies. 

Overall, the search is a great tool, but if you are used to using Google’s Shopping and Images functions, you may not want to use it as your default search engine.

I appreciate your support.