How do I operate AI agents as a data scientist in 2025

Photo by the author

# Entry

As scientists from data, we wear so many hats at work that it often seems that multiple careers have fought into one. In one business day I have to:

• Build data pipelines with SQL AND Python
• Exploit statistics for data analysis
• Give the recommendations for stakeholders
• Consistently monitor product performance and generate reports
• Take experiments to assist the company decide whether to start the product

And it’s only half of it.

Being a scientist from data is invigorating because it is one of the most versatile fields in technology: you get an exposure to so many different aspects of the company and you can visualize the influence of products on everyday users.

But minus? You always play.

If the premiere of the product works poorly, you need to find out why – and you have to do it immediately. In the meantime, if the stakeholder wants to understand the impact of starting function and instead of function B, you must quickly design an experiment and explain the results to them in an uncomplicated -to -understand way.

You can’t be too technical in your explanation, but you can’t be too vague. You need to find an environment that balances the interpretation with analytical rigor.

At the end of the work day, Sometimes I think I just ran a marathon. Only to wake up and do it all the next day. So when I have the opportunity to automate part of my work with AI, I take it.

I recently started to include AI agents in my scientific work flows.

This made me more proficient in my work and I can answer business questions with data much faster than before.

In this article, I will explain exactly how I operate AI agents to automate part of my work flow in learning data. In particular, we will examine:

• As usual, I make a flow of work in learning data without artificial intelligence
• Steps taken to automate the flow of work with AI
• The exact tools I operate and how much time it saved me

But before we get into it, let’s visit what exactly the AI agent is and why there is so much noise around them.

# What are AI agents?

AI agents are systems powered by a enormous language model (LLM) that can automatically perform tasks through planning and reasoning through the problem. They can be used to automate advanced work flows without a clear direction from the user.

This may look like launching a single command and having LLM to carry out a comprehensive workflow while making decisions and adjusting its approach throughout the process. You can operate this time to focus on other tasks without having to intervene or monitor each step.

# How do I operate AI agents to automate data learning experiments

Experiments are a huge part of data learning.

Companies such as Spotify, Google and Meta always experiment before they spend a novel product to understand:

• Whether the novel product will provide a high return on investment and is worth the resources allocated to build
• If the product has a long -term positive effect on the platform
• User’s sentiment around the launch of this product

Data scientists usually carry out A/B tests to determine the effectiveness of a novel function or product introduction. To learn more about A/B testing in data science, you can read this guide to A/B tests.

Companies can carry out up to 100 experiments a week. Designing and analyzing experiments can be a very repetitive process, which is why I decided to automate it using AI agents.

Here’s how I usually analyze the results of the experiment, a process that lasts about three days to a week:

1. Build SQL pipelines to separate A/B test data that flow from the system
2. Ask these pipelines and perform an exploration data analysis (EDA) to determine the type of statistical test for operate
3. Write Python code to start statistical tests and visualize this data
4. Generate a recommendation (for example, enter this function to 100% of our users)
5. Present this data in the form of an Excel sheet, document or slides and explain the results to stakeholders

Steps 2 and 3 are the most time consuming, because the results of the experiment are not always plain.

For example, when deciding on the introduction of video advertising or image advertising, we can get conflicting results. Ad image can generate more immediate purchases, which leads to higher compact -term revenues. However, video ads can lead to better retention of users and loyalty, which means that customers make more repetitive purchases. This leads to higher long -term revenues.

In this case, we must collect more popular data points to decide to launch image or video ads. Perhaps we will have to operate various statistical techniques and perform some simulations to see which approach is best in accordance with our business goals.

When this process is automated with the AI agent, it removes many manual interventions. We can collect AI data and conduct this deep diving analysis for us, which removes the analytical lifting of hefty, which we usually do.

Here’s what the automatic A/B test analysis with AI agent looks like:

1. I operate CursorAI editor that can access the code database and automatically save and edit the code.
2. Using the contextual protocol of the model (MCP), the cursor gains access to the data lake in which raw experiments are coming
3. The cursor then automatically builds a pipeline to process the data of the experiment and again gains access to the data lake to join it with other relevant data tables
4. After creating all the necessary pipelines, EDA performs on these tables and automatically determines the best statistical technique for analysis of test results A/B
5. Launches the selected statistical test and analyzes the output data, automatically creating a comprehensive HTML report on results in the format format, which is presented for business stakeholders

The above is a comprehensive framework for automating the experiment with AI agent.

Of course, after the end of this process, I look through the results of the analysis and pass the steps taken by the AI agent. I have to admit that this flow of work is not always trouble -free. Ai does a hallucinate and needs a lot of hints and examples of previous analyzes before he can come up with his own work flow. The “Garbage In, Garbage Out” principle is definitely valid here and I spent almost a week in running examples and building rapid files to make sure that the cursor has all the relevant information needed to run this analysis.

There were many back and many iterations before automated frames made as expected.

Now, when this agent AI works, however, I am able to dramatically shorten the time spent on the analysis of the results of A/B testing. While agent AI performs this flow, I can focus on other tasks.

This removes tasks from my plate, which makes me a slightly less busy scientist. I can also quickly present the results to stakeholders, and the shorter return time helps the entire product team in making faster decisions.

# Why do you need to learn AI agents in the field of data science

Everyone I know, every professional, somehow included artificial intelligence in their work flow. In organizations, it is top -down to make faster business decisions, introduce products faster and overtake the competition. I believe that AI adoption is crucial for scientists from data to remain current and remain competitive in this labor market.

In my experience, the creation of agency workflows to automate some of our tasks requires us to raise us. I had to learn novel tools and techniques, such as MCP configuration, monitoring the AI agent (which differs from entering the poem in Chatgpt) and work flow orchestration. The initial learning curve is worth it because it saves hours when you can automate the part of the work.

If you are a scientist or aspiring, I recommend learning how to build work flows assisted by AI at an early stage of your career. This quickly becomes industry expectations, not just love, and you should start to set up in the near future of the roles of data.

To start, you can Watch this movie To get a step -by -step guide on how to learn agentic ai for free.

Natassha Selvaraj He is a scientist of self -taught with passion for writing. Natassha writes about everything related to data, a true master of all data topics. You can connect with it LinkedIn or check it YouTube channel.