The Strange Creature Called an AI Project

The other day, an agent I had worked with before contacted me.

“There is a system development project using AI. Would you be interested?”

Apparently, they did not know the details yet.

I replied that I would at least like to hear about it.

I have been getting more of these messages lately.

But every time, I think the same thing.

What exactly is an “AI project” supposed to mean?

AI Projects Are Too Vague

For example, if someone says “Java project,” I can roughly imagine what it means.

Whether it is a web system or a business system is another question, but at least I can see what kind of person they need to build it.

With AI projects, however, the conversation suddenly becomes vague.

AI is simply too broad.

In my head, I divide it roughly into these categories.

1. Model Development

This is work that creates AI itself.

It involves researching new models or improving training algorithms.

Mathematics, statistics, paper reading, GPU clusters.

At that point, it is basically a research role.

Honestly, this is outside my main area.

I also do not really think these kinds of projects often come through freelance agents in the first place.

2. Machine Learning

This is also a broad category.

Some people design and train models. Others integrate already-built models into systems.

The former is closer to a data scientist.

The latter is closer to a software engineer.

But when I look at job descriptions, both are sometimes grouped together under “AI engineer.”

No, those are fairly different jobs.

3. LLM System Development

I think this is the most common category recently.

This means building systems using ChatGPT, Claude, Gemini, and similar tools.

It includes building RAG systems, agents, and workflows.

This is much closer to my actual work.

AWS, databases, API integrations, authentication infrastructure.

In the end, most of the work is ordinary system development.

AI is an important component, but it is not the whole system.

What I Can Actually Do

Personally, I do have a mathematical background.

I studied in the sciences, so I do not have an allergic reaction to statistics or formulas.

But I am not a machine learning researcher.

If someone asked me to implement a Transformer from scratch or train an original model, that would be difficult.

On the other hand, I can handle work such as:

• Building systems that use LLMs
• Implementing them on AWS
• API integration
• RAG
• Agent systems

These are all within the range of work I can do professionally.

So when someone says in a project explanation, “We are doing AI,” I want to ask:

“Which AI?”

I Think Clients Are Also Unsure

This is probably the interesting part.

Clients and agents still do not seem to have a good way to classify AI talent.

In the past, they could simply say:

“Hiring a Java engineer”

“Hiring an AWS infrastructure engineer”

But now it becomes:

“Hiring someone who can do AI”

Then, when I actually hear the details, it turns out to be something like:

“We want to build a search system using the ChatGPT API.”

That is not a job for an AI researcher. It is ordinary system engineering.

I Suspect This Project Is the Same

Of course, I will not know until I hear the full details.

Still, I do not really think a serious research role would be introduced through a freelance agent.

Most likely, the conversation is:

“We want to build something using AI.”

And what they actually need is:

“Someone who can build a system while using AI as one component.”

At least, that is my guess.

Even so, recent AI projects really are strange.

The project name says AI.

The job description says AI.

The interview also says AI.

But as the conversation goes on, it always ends up being about AWS, databases, and APIs.

In the end, even in the AI era, the work of engineers has not changed all that much.