MAKE Issue: Key Checkpoints for AI Agent Adoption
4 essential steps every company preparing to adopt agentic AI must know
AI is powerful. But it delivers the greatest impact when carefully planned and thoughtfully applied.
Make's VP of Market Strategy, Darin Patterson, shares four crucial points to remember when building your business around agentic AI.
Not long ago, I had the opportunity to participate as a panelist at the Hard Skill Exchange Agentic AI Summit.
During this session, we had deep discussions with various experts on how to develop a successful agentic AI strategy.
Timed with a pivotal moment in the AI industry, the conversations with leaders from different fields were extremely interesting and thought-provoking.
The panelists, including myself, carefully explored the practical steps companies need to adopt agentic automation in a procedural, safe, rational, and effective manner.
Let me share the key insights I gained here.
AI Agent Adoption Step 1
Planning the Right Infrastructure Is Most Important
Over the past 100 years, humans and automation technology have evolved together.
Thanks to this, how companies evaluate human work performance is well-established.
But with the rapid emergence of AI-first companies and the need for existing companies to adopt AI to stay competitive, new questions have arisen.
Namely, "How should we evaluate work performance in the era of agentic systems?"
Clear standards and frameworks for how to monitor and evaluate the performance of AI-integrated work processes are needed now and will be needed even more in the future.
- Organizations must think about what specific role AI agents will play at every stage
- Exactly what data they need to perform well
- And even if AI works well right now, how to detect and correct performance degradation or unexpected problems later on
These must be carefully considered.
Hyoun Park, CEO and Principal Analyst at Amalgam Insights, who moderated the panel, summarized it neatly:
"Data access, the AI model built on top of it, the agent interface configured as we want it, and determining how to manage outputs — all of these ultimately stack together like a sandwich to form a single 'AI stack.'"
Like any typical technology infrastructure, if you carefully consider and manage each component of the AI stack, it will ultimately have a significant impact on your work efficiency, scalability, and project success.
AI Agent Adoption Step 2
Quality Data — More Important Than Ever, Yet Easier to Handle
As with any stable system, the quality of input data is critical for agentic AI as well.
Good data produces good system quality, but bad data makes outcomes unpredictable.
While everyone has always agreed that data quality is important, maintaining it consistently in practice has been difficult.
You know how hard it was to ask sales staff to consistently enter data into CRM systems (like Salesforce), right?
The act of consistently entering data has been a long-standing problem across many companies.
As a result, trying to work with insufficient data repeatedly led to unsatisfactory outcomes.
But in my opinion, this is precisely one of the biggest opportunities AI brings. In traditional approaches, all data had to be collected perfectly in exactly the same format.
However, AI excels at listening to sales call recordings, writing call notes, identifying key points automatically, and leveraging these in a more natural form.
Therefore, the effort required to collect data in perfect formats will decrease compared to the past and can be more easily improved.
With AI, data can be collected and recorded in a relatively nuanced and rich format for practical use.
AI Agent Adoption Step 3
Considerations for Ethics, Governance, and Access Control
At the same time, you must also think about who can access what parts of the system to get the data you need, and how transparently everything should be operated.
One area I think about frequently is "Should AI clearly identify itself as AI?"
For example, if AI is sending sales or marketing messages, regulations are likely to increase going forward.
In the U.S., the FCC (Federal Communications Commission) is already addressing this issue.
Whether AI must identify itself as AI going forward, and whether customers need to consent to AI communication — while there are no clear standards at present, this is an issue every company must watch and respond to.
Similar considerations are needed within organizations as well.
In a few years, you might join a new company, have a Zoom meeting with several colleagues, and later discover that one of them was actually AI, not a real person.
While this is hypothetical for now, it will soon be an issue companies need to clarify.
Whatever tasks you delegate, you must carefully consider exactly which systems and data AI is allowed to access.
No matter how capable AI is, if you let it have unrestricted access to your entire CRM data, trust erodes, legal problems can arise, and you might encounter serious issues you never anticipated.
Therefore, a clear approach to limiting and managing the systems and data that AI agents can use is essential.
Agentic AI is a tool that can bring truly enormous possibilities. However, simply adopting it isn't the end — you must carefully follow four practical steps.
If you establish the right infrastructure, meticulously manage the data AI uses, and properly consider ethical issues and access controls, your company can leverage AI efficiently and successfully.
AI Agent Adoption Step 4
AI Models, Understanding Your Work, and Building Agent Interfaces
It goes without saying, but the most important thing in establishing this approach is truly understanding your company's work processes.
To apply automation successfully, you need to know exactly how things work.
When trying to adopt AI, this understanding needs to be even more thorough.
Yet even if you know your company's operations inside and out, you often can't pinpoint exactly where AI will have the greatest impact.
Ultimately, you have to find out through direct experimentation. To experiment well, you need a solid foundation first.
In other words, you need a kind of "model framework" that helps you think about what resources are currently available and how to best utilize them.
The first foundation to build isn't particularly difficult, but it requires consistency.
It's about understanding the strengths and weaknesses of various AI models well.
There's no perfect super-model among the AI available today that excels at everything.
And one won't appear for a while. Ultimately, it's important to keep learning and staying up to date.
Second, your business model itself should be built to allow small experiments, testing, and iterative improvement.
It's about training each employee to try new things, fail often, and gradually create better value.
This approach is actually one of the secrets behind how today's rapidly growing "ultra-lean teams" reach $100 million in annual revenue.
Above all, the most important thing is giving team members the right tools to easily put these small experiments and iterations into action.
You need to create an environment that's easy to use and flexible enough to attach and detach any AI model.
So whether someone can code or not, everyone should be able to fully enjoy the benefits of work automation.
Simply put, aligning your company's knowledge base, organizational culture, and AI tools with the principle that "anyone can easily try various things and gradually improve" is the way to adapt well to a changing era and achieve better results.
Management Should Be Visual and Easy for Everyone to See
This will naturally create new skill requirements even for traditionally defined job roles.
Employees will also need to understand how AI operates within their workflow and will have more tasks involving work design and management.
Ultimately, understanding how tools and AI agents are connected through inputs and outputs becomes important.
Going back to the original point, how should we make performance and results easy to see and evaluate?
I believe the most efficient approach is to visualize as much as possible so you can see things at a glance.
That's one of the reasons I became drawn to Make's approach.
Over the past few years, I've been consistently interested in how to visually represent automation within existing workflows and processes.
When you can visually see how data flows, it becomes easy to understand workflows, quickly identify where problems lie, and find solutions much more conveniently.
But as more and more AI agents emerge, I'm thinking about how to visualize an even more complex situation.
We need to easily visualize where existing automation is performing well, where AI needs to be added for more power, whether new tools for AI are needed, whether tool outputs need further refinement, and so on.
We also need to think about how to easily view and manage the interactions between specialized agents as they collaborate. After all, the more complex things get, the more everything needs to work together harmoniously.
Ultimately, as AI becomes more deeply embedded in our work, the key phrase will be "orchestration and management." Going forward, more people with diverse technical skill levels will be managing work that involves AI.
These people need to be able to manage visually, seeing things directly with their eyes. Of course, the actual operations may happen behind the scenes.
But from the viewer's perspective, everything should be easy to understand at a glance. This kind of visual management will be essential going forward.
That way, the AI tools your company uses can deliver the most effective results across the entire organization, quickly and reliably.
Building an AI-Centric 'Tech Stack'
I agree with summit moderator Hyoun Park's opinion. Organizing the concept of an AI tech stack will be very important for successfully applying AI.
Putting Hyoun Park's thoughts in my own words, data, accessibility, interface, and orchestration/management are the four essential elements that every company needs as it builds out its AI capabilities.
As agentic systems grow and business methods change rapidly, it will be important for every company to use these four elements as a foundational framework going forward.
IMPAKERS Blog | Make Custom Function Tips! Read more
Source: Darin Patterson, Make, "4 steps every business must consider when building your agentic AI stack", https://www.make.com/en/blog/building-an-agentic-ai-stack, (2025.06.03)