Artificial intelligence is changing how we work, learn, and solve problems. Software that once took weeks to develop can now be built much faster with AI assistance. For Bangladesh, this creates opportunities to strengthen technology exports, improve services, and build more sophisticated products.
Yet through my work with technology professionals in Bangladesh, I have noticed something worth examining. People using the same AI tools often achieve very different results. Some define problems clearly, question responses, and improve the outcome. Others accept recommendations without fully understanding them.
Technical experience matters. But I have increasingly observed that English comprehension, reading comprehension, logical reasoning, and critical thinking may also influence how effectively people use AI. This is my professional observation, not a scientific conclusion. It raises a question: are we focusing too much on access to AI and too little on the skills needed to use it intelligently?
AI adoption discussions often focus on infrastructure, computing resources, and training. These matter, but access and capability are different. Anyone can ask AI to write software, prepare a financial analysis, or recommend a business strategy. The result may be polished and convincing. But can the user understand its assumptions, recognise errors, and decide whether the recommendation is appropriate?
AI makes sophisticated work easier to produce, not necessarily easier to understand. This distinction matters when outputs influence financial decisions, security, or public services.
Bangla is our mother tongue, and English proficiency should never be confused with intelligence. Someone who reasons effectively in Bangla is not inherently less capable than someone fluent in English.
Still, much technical documentation, scientific research, and programming material is published in English. Language can also influence AI performance. A 2024 study, BEnQA: A Question Answering Benchmark for Bengali and English, evaluated approximately 5,000 parallel Bengali and English questions relevant to Bangladesh. It found performance differences between languages and that English translations can help in certain settings. Another 2024 study, BenLLM-Eval, identified limitations in large language models’ Bengali capabilities. Neither study proves that English always produces better results. Performance depends on the model and task, and Bangla can be preferable for local context. Nevertheless, English comprehension may offer Bangladeshi technical professionals a practical advantage. It is only one part of the picture.
Reading a technical explanation is not the same as understanding its implications. Someone may recognise the words yet miss an important qualification or risk. Consider an engineer using AI to develop an application. The AI reports that all existing tests passed but also identifies potential security vulnerabilities and untested scenarios. One engineer concludes that the application is ready. Another investigates the unresolved risks. Both received the same information. Their comprehension and judgment led to different decisions.
Similar problems arise when an analyst reviews an AI-generated forecast or a manager evaluates a business recommendation. A convincing answer is not necessarily a verified one.
Critical thinking matters for the same reason. An AI assistant may recommend software architecture suitable for a demonstration but unsuitable for thousands of users. An engineer must question its scalability, reliability, and security. Effective AI use requires knowing when to trust a response and when to seek evidence.
AI is increasing our ability to produce sophisticated work. Our ability to verify that work may not be improving as quickly. I think of this as the AI verification gap. A developer can generate thousands of lines of code without understanding the architecture well enough to identify failures. A business professional can produce a market assessment in minutes without knowing whether its sources or assumptions are reliable.
The concern is not that AI necessarily produces poor work. Often it produces excellent work. The concern is that the sophistication of the output may exceed the user’s ability to evaluate it, a term used as a framework for discussing this risk, not as an established scientific measure. Organisations need people and processes capable of meaningful oversight, particularly for consequential decisions.
We should not assume that AI only benefits highly skilled professionals. In Generative AI at Work, Erik Brynjolfsson, Danielle Li, and Lindsey Raymond studied more than 5,000 customer support agents. AI assistance improved productivity by approximately 14 percent on average, with particularly large gains among less experienced workers. That is encouraging. AI can help people overcome limitations in knowledge and experience. Yet it may also reduce differences in routine execution while leaving significant differences in complex problem-solving and judgment. Both effects could exist simultaneously. We need evidence rather than assumptions about whether AI narrows or widens particular skill gaps.
Bangladesh has capable engineers, researchers, and entrepreneurs. The question is whether our education and training systems develop comprehension and reasoning consistently, and at sufficient scale. Concerns have long been raised about excessive memorisation and examination-focused learning in parts of our education system. Are students learning to interpret complex material, distinguish evidence from opinion, challenge assumptions, and explain their conclusions?
In my experience working with engineering teams serving both Bangladeshi and international clients, the strongest contributors are not necessarily those who type code fastest. They are often the people who understand the underlying business problem, ask clarifying questions, recognise what has been left unsaid, and take ownership of the result. AI makes this distinction more visible because producing a first draft of the solution is no longer the hardest part.
This is both an economic and an educational issue. Moving from lower-cost software services toward higher-value engineering and Bangladeshi-owned products requires professionals who can understand customer problems, evaluate alternatives, and take responsibility for complex solutions. Employers may need to rethink recruitment. Programming knowledge remains important, but routine code generation is becoming easier. Instead of asking candidates only to write code, we could give them an AI-generated solution with weaknesses and ask them to identify the problems and explain their consequences. Similar exercises could help evaluate analysts and managers.
Bangladesh should also invest in better Bangla AI, including high-quality datasets, evaluation benchmarks, and applications that understand local institutions and terminology. People should not need exceptional English proficiency to benefit from advanced technology. Universities and training institutions could begin by incorporating exercises in which students critique AI-generated explanations, compare competing solutions, and justify their choices. These activities would test understanding rather than the ability to reproduce a polished answer.
Finally, we need Bangladesh-specific research. A study could assess reading comprehension, English proficiency, reasoning, and critical thinking, then compare participants’ performance on the same AI-assisted tasks in Bangla and English. It should account for education, experience, and prior AI exposure. Observed correlations alone would not establish causation; controlled studies could help identify what actually improves performance.
Our AI ambitions need infrastructure and technical skills, but also stronger comprehension, reasoning, and communication. AI makes answers easier to produce. Bangladesh’s success will depend increasingly on whether we understand those answers, know when to question them, and recognise when they are wrong.
The views expressed in this article are solely those of the author
The writer is a technology entrepreneur, ICT strategist, and Founder and Managing Director of Periscope Labs. Email: [email protected]
