Sapient Slingshot wins Technology Breakthrough of the Year.
Sapient Bodhi ranked #1 for deep research.
Meet Sapient Sustain: context-aware AI for complex IT operations.
3M lines of COBOL → clear specs in 8 weeks.
October 13, 2026
Loading AI-generated summary...
More than 20 years ago, I was trying to find useful signals buried inside noise. I was studying electrical engineering and working with signal data in MATLAB. Most of what I was looking at wasn’t relevant, but somewhere inside it was the information I actually needed. That was when I started to understand the importance of data.
I didn’t know then that data engineering would become my career. But since graduating, I have worked with many different kinds of data, from financial transactions to healthcare information. Every domain brings different questions about scale, sensitivity and security.
Today, as a manager of data engineering at Publicis Sapient in Toronto, those questions are still part of my work. The difference is that now I have AI helping me answer them faster.
A challenge I see often is that organizations have systems and data environments built many years ago. Their data may still sit in an on-premises SQL database, a mainframe or another legacy environment. Those systems may have worked well when they were introduced, but the amount of data organizations need to process has changed dramatically.
The expectations have changed too. Teams want faster processing. They want to reduce infrastructure and maintenance costs while taking advantage of cloud environments. And they don't want modernization to take years. That creates an engineering problem. How do you move the existing logic into a tech stack that can handle what the business needs now?
On one modernization project, I was working with legacy data-processing code that I needed to translate into a cluster-based parallel processing environment. I used our AI product, Sapient Slingshot, to help with that translation. I defined the prompts and used the tool to convert the legacy processing logic into an equivalent PySpark data-processing module. That gave me a much faster starting point than translating everything manually. But it did not mean the work was finished. There were still areas I needed to refine, including how credentials were handled through a secrets manager. I still had to understand the requirements, validate the code and make sure the solution worked correctly in the environment where it would actually run. That is an important distinction for me. AI can accelerate the engineering, but it does not remove the need for a human engineer.
I know how different this kind of work feels because I remember doing it before these tools existed.
About a decade ago, I had to build an algorithm around customer-service transactions. Roughly half a million transactions were coming in every four hours, and I needed to use them to reconstruct the lifecycle of individual cases and measure productivity accurately. The difficult part was handling all the exceptions. What happened if someone switched from one case to another, or a case stayed open overnight? If someone stopped working on a case, went home and returned eight hours later, I obviously could not count those eight hours as productive time. Every edge case affected the algorithm.
I remember going back to the same piece of logic again and again. There were multiple rounds of testing and refinement because the result had to be accurate. It ultimately took me roughly three months to get the algorithm where it needed to be.
When I look back at that work now, I think about how differently I could approach it with AI. With tools like the ones available today, I believe much of that development and testing could have been completed within days rather than months. Across the engineering work I do now, I estimate AI can reduce the effort involved in some tasks by roughly 60 to 80 percent. But there is an important condition attached to that number: You still have to check the answer.
One thing AI has taught me is not to confuse confidence with accuracy. I am currently preparing for a professional data engineering certification. Sometimes I come across a difficult question and give it to an AI tool to see how it approaches the problem. It will give me an answer. Then I might ask, “Are you completely sure? I think option C may be more accurate than option A.” Sometimes it responds by agreeing with me and changing the answer. That immediately makes me wonder, “What about the previous questions where I did not challenge it?”
It is funny, but it is also a serious engineering point. If you are dealing with sensitive financial information or another situation where a small distinction changes the outcome, you cannot simply trust an answer because the AI produced it quickly. There are parts of the work where human intervention is essential. You need to understand the data. You need to understand the business requirement. And you need to recognize when something looks technically correct but contextually wrong. That judgment still belongs to us as humans.
The other thing I find exciting about AI is how much it expands what I can learn while I am working. A project may start with Python, but that doesn't mean Python is the only thing I need. In my current work, I have used AI to help me understand Amazon Web Services Simple Queue Service messaging patterns, including what the messages look like and how to decode them. I have also used it to help build Terraform code structures that are part of the wider solution. That means I do not have to stay inside the technologies I already know. If you want to understand React, you can use it to learn React. If you inherit an unfamiliar module, you can use AI to dissect it, understand how it works and start getting better at it.
For me, the most important skill is understanding what you are actually trying to accomplish. You need to understand the functionality, the requirements and what the final outcome should be. Then AI can help you move faster.
AI has changed how I write code, investigate unfamiliar technologies and approach engineering problems. But it hasn't changed my responsibility for the result. That is what makes this kind of work at Publicis Sapient interesting to me. You can use AI to translate legacy logic, explore another technology or get to a first version far faster than before. Then you test it. Challenge it. Find the edge cases. Protect the data. Make sure it works in the real environment. AI has made it easier for me to take on a wider range of problems. But the engineer still has to decide when the answer is actually ready to ship.
