Are They Gonna Take 'er Jobs???

For a bit now I have been saying:

"AI isn't coming to take your jobs. People who can USE AI will be."

This is still very true. But "Meyer", you say, "The AI agents can do just like a hooman!!!" Well... not really. There are still some very very very large gaps that the human must fill. If you work with AI and LLMs at all you can tell in the purely generated AI work product that there is room to improve.

Firstly, AI does an excellent job at things that are cut and dried. You build an agent that will respond to customer service emails. Using the grounded data, your agent will answer most of the questions and respond appropriately to the sentiment and the asks of the customer. HOWEVER, the agent really isn't very good at Customer Service.

I choose customer service because it is an area that is fertile ground for AI automation. But, AI is a robot. It will act robotically, following the rules 100%. No grey area. No wiggle room.

How many of us have an example where we were in a bar, or a restaurant, and we got something on the house. They brought us an extra appetizer, or maybe the bartender was working on a new cocktail and you got to try it. OR... Someone in customer service understood your problem and made it their mission in life to get it solved for you. AI doesn't do that. People do.

AI, even reasoning models can't see beyond its instructions, to give you a personal touch.

That brings up the second issue... For the foreseeable future, trust issues will always be there. I think that an AI "doctor" could likely diagnose and provide me with my basic health care. I'm fairly fit, and aside from the hereditary Gout, I really don't have much cause to visit the doctor.

However... when I go see the doctor, I don't think I would trust an AI. I know better than most that the AI doctor is going to have more knowledge at their immediate disposal than any human. I also know that the "feeling" that an experienced human gets in the noble art of medicine is better than any AI working through an index of symptoms.

When your life is on the line, would we trust AI? Would you hire an AI lawyer to defend your innocence in a murder trial? I wouldn't trust it.

Finally... And most frustratingly for those of us working in the field, AIs absolutely SUCK at choosing tools. They always want to use their internal trained knowledge.

An example is this... Agents freaking love Python. Even when you say, this needs to be optimized for this kind of environment, when left on their own, they won't choose C, they will choose Python.

They also have a very tough time in the use of design patterns that optimize performance. It isn't that they don't know what these patterns are, they just don't seem to make the connection that in XYZ situation, you use ZYX pattern.

What I normally use as my baseline ask for AI is the Singleton pattern. Common enough right? Non coders stay with me! A Singleton is where you need to ensure that a class only has ONE instance. (Like the government... there is only one US government, despite the different people and personalities, there is only ONE. We can't spin up another US government. They tried that in 1861... Didn't go well. But I'm seriously digressing now.)

I come up with a situational description that would be solved by a singleton. Most of the time it has to do with data classes. I ask the AI to create the code, using the most optimized and best language for the case.

I have yet to get a singleton pattern. And it is ALWAYS in Python.

Coming back to the trust issue, I can't trust the AI with my health care, when I can't even trust it to pick the best design pattern and language for the task.

Bringing this all home... There MUST be a human in the loop. MUST. Who is that human? YOU! Because, you are human who is interested in and using AI.

Who is out? The person who is unwilling to BE the human in the loop. The person who refuses to use the AI in the cut and dried situations. The person who doesn't leverage AI as the greatest tool to have ever been invented in the history of mankind (yeah seriously).