Bangladeshi educator promotes human-led AI coding loop for learners

As artificial intelligence (AI) tools become ubiquitous in software development, Bangladeshi software engineer and educator Sumit Saha has emerged as a leading voice advocating for a "fundamentals-first" approach to programming.

Sumit, the founder of the popular platform 'Learn with Sumit', argues that beginners must learn to read, reason about, and debug code independently before relying on Large Language Models (LLMs) like ChatGPT or GitHub Copilot. His platform, which provides Bangla-language programming education, has reached over 173,000 YouTube subscribers and a community of nearly 100,000 members by 2025.

While tools like Claude and AI coding agents accelerate development, Saha warns that speed should not replace engineering judgment. 

He highlights a common vulnerability in AI-generated e-commerce systems: an AI might create a functional-looking shopping cart that only applies discounts on the client side. Without server-side validation, a user could manipulate the browser to complete a purchase at a fraudulent price. This risk aligns with the Open Worldwide Application Security Project (OWASP) guidelines, which state that applications must never rely solely on browser-side controls for business logic.

To bridge the gap between automation and understanding, Saha introduced the "AI coding loop" in a freeCodeCamp article. This framework positions the human developer as the final decision-maker who reviews logic and tests assumptions, rather than a passive recipient of generated syntax. "Building great software isn’t about perfect prompts; it’s about a disciplined process," Saha writes, emphasizing edge cases and safety rules.

The 2025 Stack Overflow Developer Survey provides a backdrop for this caution. While 84% of developers now use or plan to use AI tools, 46% expressed distrust in the accuracy of AI output, with only 3% reporting high confidence.

Saha’s stance found support at Stanford’s 2026 AI+Education Summit. Professor Mehran Sahami, Tencent Chair in Computer Science at Stanford, noted that the challenge is to treat generative AI as a topic of study rather than a mere shortcut. "Our challenge is to... build a curriculum in which we teach how to use it to foster things like creativity and deeper educational outcomes," Sahami said.

Despite his warnings, Saha does not reject AI. Through Logicbase Labs, he demonstrates advanced workflows, such as building secure AI pull-request reviewers using Claude and GitHub Actions. His collaboration with freeCodeCamp has expanded his reach to a global audience of over 15 million developers, with his technical handbooks featured in newsletters reaching millions.

For Saha, the path forward is clear: novices must master the basics first to ensure they remain in control of the technology, rather than becoming dependent on unverified automation.