If you believe the posts in your social media feed, everyone is building agents all the time that are doing all the work…sort of. I see that vision, too. But after spending the last two months setting up, tweaking, replatforming, and expanding my own team of agents, I also see three main hurdles: evolving AI literacy (yes, still!), making products easy to use, and empowering users to fill the gaps. And then there's the obvious one: getting AI labs to stop skewing the debate toward existential threats.
I will be attending Gartner HR Symposium/Xpo™ in Orlando, FL from October 26-28, 2026, to talk to HR leaders, software providers, and market analysts about how they are addressing the hybrid workforce of humans and agents. Use code GartnerHR to save $600 on your ticket.
AI Literacy Evolves at the Pace of Technology
Two years ago, AI literacy meant knowing different prompting frameworks and techniques to get ChatGPT to perform well, and having a basic awareness that AI hallucinates. Since then, AI labs have greatly improved their models, and those improvements have been folded into the models themselves as reasoning models. After all, most of these early techniques were workarounds for technology shortcomings. Personally, I don’t recall the last time I explicitly told Claude, “You are an experienced … with 25 years of experience …,” because AI tools have improved so much. AI literacy has evolved to asking yourself if Claude or Copilot could support you on a task (and finding out that it does).
What to do: Organizations must continue investing in AI upskilling to move the baseline of knowledge and literacy at the pace of technology. Individuals need to continue exploring and testing the boundaries of what AI can help with. Lastly, IT teams supporting the business need to provide state-of-the-art tools that enable innovation while ensuring governance and compliance.
New LinkedIn Learning Course: Lovable 101: From Prompt to Product
Building a business app used to require a development team, a budget, and months of work. Today, AI-assisted tools like Lovable make it possible for any business professional to go from idea to published app in under an hour with no coding experience required.
In this course, instructor Andreas Welsch walks you through the complete process of building a Wins Journal app, a personal tool for logging professional achievements and generating an AI-powered year-end summary.
Along the way, learn how to validate an app idea, write effective prompts, add user authentication, store personal data securely, and publish your finished app. Walk away with a working, published app and the confidence to build your next one.
(Click the image to start the course on LinkedIn Learning…)
Building Agents Becomes Easier
Tech leaders have been promoting the idea that everyone will have their own team of agents they can delegate to. I am confident that will be the case long term. But it requires making the technology as easy, seamless, and foolproof as possible, such that anyone can use it, no matter their level of technical expertise. This could involve answering a few questions about yourself, providing the agents with your data (live or exported), deciding which roles you need besides a chief of staff and email triage, and dynamically configuring or building them. (Hint: In a business environment, it will most likely be how to get agents to draft your LinkedIn posts, like social media influencers often suggest. Instead, it will be financial forecasting, reconciling payments, finding discrepancies and addressing them, and strategizing “what if.”)
Sure, you can build your agents in products like OpenClaw, Hermes, Claude, and others. Users who want to set up an agentic team need to get familiar with these platforms or build everything from scratch. This is why AI literacy and exploring the possibilities are prerequisites. All of this should (and will) be built into business apps over time.
What to do: The building blocks for agentic systems exist, but it takes technological savviness to create your own team of agents, avoid security pitfalls, and achieve reliable results. For AI agents to become mainstream, software vendors will need to make them disappear. You will recognize this next level of value when AI agents become a common technology like website builders, digital payments, or product recommendations, and the average user doesn't have to think about the stack underneath (aka ambient intelligence).
Thanks for reading The AI MEMO! This post is public, so feel free to share it.
As AI-ready users look to expand the scope of tasks their agentic team covers, they quickly hit several roadblocks. Tools and applications users can access in their browser aren't readily available to the AI agent. Claude and others now ship with an integrated browser that navigates websites on a user’s behalf. This is a workaround for some of these limitations of the web experience built for humans. Although built-in browsers make websites accessible by agents, these tools lack the governance, observability, and reporting needed to use them as reliable automations.
But that’s not all. Worst case, the app that's a key part of your process doesn’t have an API. Best case, its API isn't available via an MCP Server (which lets AI agents interface with external apps and data). In your personal AI setup, building and deploying an MCP server is manageable, even with limited technical expertise. Claude and others not only generate the code, but they will also walk you through the steps to secure it and set it up. In a business environment, IT quickly becomes the gatekeeper and bottleneck.
Organizations clearly need governance and oversight for AI, including who has access to which systems and data. But when users who could automate a process keep running into red tape that slows them down, Agentic AI scale takes a hit. If humans need to do manual integration, a patchwork remains, and the full potential stays untapped. If users are determined to automate a task, they will look for ways to make it happen (and create shadow AI in the process).
Bonus: Maintaining Agentic Teams and Governance
Assume for a minute that these first three layers are resolved: users are proficient enough to determine when to use AI and what to use it for, it’s just a few clicks to set up your agents, and any workflow gaps are closed. Then, operations is next.
That’s where organizational principles come into play. Like human team members, agents need to store different kinds of information in different places: their own instructions and memory, their current state, and information shared with others. New information needs to be captured, evaluated, and promoted from signal to knowledge.
But when does that learning happen, and how does communication take place? Applying software development lifecycle principles can effectively maintain your agentic team. In a human team or organization, this process is equivalent to one-on-one check-ins, team meetings, training, and updating standard operating procedures. Like a leader, the agentic team operator reviews and approves this learning loop.
What to do: Dedicate one agent on your team to reviewing and maintaining the entire team setup. Run a weekly maintenance review that surfaces duplicate, inconsistent, and insecure instructions, and a specification for how to fix them. Review the proposal and apply what makes sense.
Summary
Agentic AI promises to unlock a new wave of benefits and value for both businesses and consumers. But for agents to become mainstream, the tech industry needs to remove several present-day hurdles and speed bumps. AI literacy remains a critical and evolving building block in any talent development strategy. Agentic products and services need to become easier to use. IT departments must keep balancing rapid innovation with governance and security. If Agentic AI technology and tooling mature, business users and consumers will benefit as well, beyond technologists.
Become an AI Leader
Join my bi-weekly live stream and podcast for leaders and hands-on practitioners. Each episode features a different guest who shares their AI journey and actionable insights. Learn from your peers how you can lead artificial intelligence, generative AI, agentic AI, and automation in business with confidence.
October 13 - Michael Nevski (Consumer Market Intelligence Expert) will discuss the prerequisites for AI agents to complete financial transactions on their users’ behalf. [More details to follow on my LinkedIn profile…]
Watch the latest episodes or listen to the “What’s the BUZZ?—AI in Business” podcast
I don't know what it's like over the in the US, but from my personal experience with mid-sized and larger companies in Europe, the main obstalce in adoption is a worry that the agent will 'misbehave' in a way that has lasting and negative consequences. I think their worries are usually exaggerated by an order of magnitude but that's the sentiment here
I don't know what it's like over the in the US, but from my personal experience with mid-sized and larger companies in Europe, the main obstalce in adoption is a worry that the agent will 'misbehave' in a way that has lasting and negative consequences. I think their worries are usually exaggerated by an order of magnitude but that's the sentiment here
An invisible agent still needs very visible limits.