The top five things you should know about agents on the Microsoft platforms.

AI Technology is Capable

I think many of us are aware that large consulting and tech companies will make announcements of products and capabilities before they are truly viable for a business. Sort of like academics believing something to be great, but there is no way to deploy that process or approach in business. While that has proven true for AI agents, we are starting to catch up and the technology certainly is capable.

The tools and proven techniques are two areas that need to catch up now so that we use the AI in the right way.

More Improvements to Come

Microsoft has devoted most of their programmers towards AI: building AI into their existing products and creating new AI products. Microsoft already has a great suite of AI tools, infrastructure, and products. Let’s imagine for a moment what would happen when a tech giant focuses on doubling that investment…I’m betting on Microsoft being one of those leaders. Microsoft has started communicating that Microsoft Dynamics 365 is an “AI ERP,” which is their way to start educating the market that soon traditional ERPs will be considered the mainframes of the 2020s. Dynamics customers have piece of mind however because their cloud ERP will stay up to speed and grow into the AI ERP of the future, without costly reimplementation.

Realistic Expectations

When calculators were released, people were concerned that humans would lose some of their math skills, or problem solving skills. The truth is that calculators let us stop spending so much time on the calculations and focus our brains on other areas. AI is similar, we can and will automate the things we can so that human brains can be focused on other tasks. When we gave people calculators, printers, and laptops we didn’t fire humans; our efficiency went up and more was getting done.

Having said that, large language models as we have them today are not as skilled as humans in all areas. I still think of these large language models as an intern who just graduated. Lot’s of head knowledge, little experience, requiring clear instruction and supervision. Many companies are very focused on clear instructions, but fewer are putting the proper focus on monitoring, supervising, and improving their AI solutions.

Tools and Focus

A common trap is thinking that AI can handle all your information, and make the right decisions every time. It is better to focus the LLM on a specific task, providing the tools, data, and instructions that pertain just to that one task. A smart idea is to ask the AI why it made a certain decision, or have a second AI grade the first AI’s answer before taking action. You need to provide the AI automation a combination of call scripts (deterministic work) and some nondeterministic (which is the best choice?). Deterministic automations are easier to test and validate, but even nondeterministic work can be verified, by using AI to grade AI.

I’m over the blogs and videos talking about 20 tools to accomplish one task. I aim to provide you real understanding of what is going on and what is real. What did I miss?

Original Post https://calafell.me/top-five-things-you-should-know-about-ai-agents/