AI Accelerator: Anas Cheriya, Principal Software Engineer

In this installment, we talk to Anas Cheriya, Principal Software Engineer, who has spent years building a hands-on relationship with AI, from early skepticism to architecting multi-agent systems in his own time. He’s embedded AI into the way his team develops software and is watching closely to see where the technology goes next. Here’s how his thinking has evolved, and where he still thinks humans need to stay firmly in the loop.

Q: How has your relationship with AI evolved to where it is now?

A: I first started using AI when ChatGPT was released in 2022. Initially, I was fairly cautious. The early tools had clear limitations including hallucinations, and I was skeptical because I wasn’t getting the output I expected. With more advanced models, the results have improved significantly. The real turning point for me was when AI agents began to take control of execution workflows.

That shift made me rethink my approach and adopt an AI-first mindset for problem solving. Over time, I invested in learning and refining prompt engineering, which became my primary focus as large language models (LLMs) evolved. AI can still hallucinate, of course, especially without proper context. I still believe AI can make mistakes in complex scenarios, so continue with what I consider a healthy level of caution, keeping the limitations in mind.

Q: How has AI become part of the way you work and experiment outside work?

A: When Model Context Protocol (MCP) was released I built an MCP server for Confluence’s PARis solution along with a user interface to interact with it. At that time, it was still an early-stage initiative but like so much with AI, it has progressed rapidly.  

AI is now fully integrated into the PARis software development process, and I have created skills and guidelines that help the team identify issues, support and analysis.

AI has become my pair programmer across writing code, designing, debugging, reviewing, and other software development lifecycle workflows. Time spent doing tasks such as searching through articles, reading Stack Overflow, and trial-and-error coding has drastically reduced. Overall, it has made my work more efficient and frankly, more enjoyable!

In my personal time, I’ve been working on a multi-agent solution for the past two to three years. The concept is to enable AI to create and manage other AI agents, and it has matured significantly recently. This hands-on experimentation outside work has really accelerated my knowledge of an AI-driven development strategy, which I can then bring to my work environment.

Q: You previously mentioned efficiency. How was AI made you more efficient?

A: Previously, I would spend hours setting up codebases, drawing architectural diagrams, and learning new technologies. I would spend days understanding system design across different use cases. This is how developers worked! Now, AI makes this analysis much faster and helps me make design decisions with a clear understanding of trade-offs.

AI-assisted coding in particular has been a major productivity boost. And I am happy to say AI has significantly increased my enjoyment of the profession. Previously, I had to read 5–10 articles to understand trade-offs between solutions, which is no longer necessary. Today most of the information I need is accessible through a single prompt.

AI has significantly increased my enjoyment of the profession.

Anas Cheriya, Principal Software Engineer.

Q: Where do you see limitations to what AI can do?

A: I think for most developers, the biggest frustration is that AI does still make mistakes. When working on complex problems, I always need to double-check responses. One strategy I use is asking AI to act as a “devil’s advocate” to validate results. Another limitation is context handling. AI has context limits, so I need to carefully provide the right input. I don’t fully trust AI to always pick the most relevant information. The more precise the input, the better the output.

The lack of true Artificial General Intelligence (AGI) remains a major limitation, which ultimately means there is still a pivotal role for humans.

The more precise the input, the better the output.

Anas Cheriya, Principal Software Engineer.

Q: What is next for AI and for the software engineering profession?

A: I’ve been following developments around models like Mythos and Fable. Mythos appears to be very powerful and potentially risky if misused. I’m interested in seeing whether it meets expectations related to AGI capabilities.

I see software engineering moving further toward AI-driven development, where humans remain in the loop to approve and refine AI-generated outputs. My advice to new graduates would be to focus on prompt engineering and learn how to work effectively within AI-driven development workflows. They should also understand guardrails and the limitations of AI. In addition, building strong system design fundamentals is essential for long-term success.