Every company we’ve consulted with lately is worried about how AI is affecting their valuation and perception in the market. Companies that aren’t using AI are viewed as falling behind, yet many are apprehensive about how to use it in a way that makes sense while managing security effectively. AI can miss the mark - especially if you use AI that’s off-the-shelf.

First let’s clarify the terms – because AI isn’t just one thing and each ‘layer of AI’ builds on the previous layer.

Machine Learning – layer one contains models trained on datasets that have been categorized and indexed. These models, including GPT, allow us to ask questions about the information and get a quick response. Think of data.

Agentic AI – layer two is where agents begin to take your requests and store them as repeatable tasks for automation. Think of repeatable tasks.

Generative AI – layer three takes all the information in the previous two layers to create something new – or generates it. Think of a creative layer that brings it all together.

Article content — Making AI Make Sense: Layers of AI

To bring this into the real world, let’s take an example of a company with lots of professional development video and text lesson content created over decades.

The first layer we would build would leverage machine learning. The prompt would be something like “Watch and read all of this content for me, create summaries, and categories for the lessons.” This allows us to ask for specific content like a list of videos that could be training for a sales team, as an example.

The second layer would make connections with repeatable tasks. The idea here is that an agent could “take all the data and assemble a lesson plan for a junior sales person.” The agent takes all the information and creates a list of content that is relevant based on the criteria.

Finally, Generative AI is when you would ask the system to “create charts, graphs, and images" for the lesson plan and add in the latest research on the subject to ensure its current.

Security and Vendor Vendor Concerns

No matter how you look at it, companies need to be selective about what data is sent out from the company and what data comes in to the company. This can be complex to navigate when so many off-the-shelf products promise to make your life easy and keep your data secure. What security settings will result in what outcomes? Is this setting really the right level of security for your data?

When we work with companies on AI projects, we always put guardrails around it. This protects your data, and ensures industry standard security practices are in place.

Another issue with the many off-the-shelf solutions is the risk that these AI companies will suddenly go out of business and leave you with nothing. If you have a customized AI solution, you can quickly pivot from one service to another, or turn off a feature completely if there is an issue. You manage the evolution and control the outcome.

The Real Focus

Knowing what you actually want to accomplish is the real focus. You need to be able to understand and articulate the outcome you are seeking before you ask AI. This means you need software and product creation expertise to know how it all comes together technically and functionally – because each AI model has strengths and weaknesses. There is no one-size-fits-all AI solution.

Why Off-the-Shelf AI Misses the Mark