An early chapter in Bangladesh’s garment-export story offers a useful lesson for its artificial intelligence ambitions. The partnership between Desh Garments and South Korea’s Daewoo in the late 1970s involved more than access to machinery. As the World Bank recounts in its 2025 report, Frontier Firms and Job Creation in Bangladesh, Desh provided land and equipment, while Daewoo financed training for Bangladeshi supervisors and managers in operating a plant, maintaining quality, and reducing material waste. Technology transfer included learning how to organise production, not merely acquiring the means to produce.
That distinction deserves renewed attention. A sewing machine and an AI model are very different technologies, but acquiring either is not the same as knowing how to build a productive organisation around it. For Bangladeshi organisations already experimenting with AI, the immediate challenge is therefore not necessarily finding a more intelligent model. It is turning the intelligence available to them into dependable work.
A revealing contrast appears in PwC’s 2026 Bangladesh CEO Survey. Most respondents said their organisational culture supported AI adoption. Yet just over four in ten reported having a clearly defined AI roadmap, while just over one in five considered their investment sufficient to achieve their AI goals. Based on responses from 45 CEOs surveyed in late 2025, the findings cannot represent every business in the country. They nevertheless suggest a gap between enthusiasm and organisational preparation. An employee using AI to finish a report faster has achieved something useful. But whether that report produces a better decision depends on the organisation receiving it. Does the information reach the right person? Is it checked? Can anybody act on it? An individual’s time saving is a real gain; it is not, by itself, evidence of institutional change.
Consider a hypothetical garment exporter whose merchandising team uses AI to answer buyers more quickly. The benefit is limited if production records are outdated, delivery commitments cannot be confirmed and revised instructions do not reach the factory floor. The company might communicate faster without becoming more reliable. For an exporter, the meaningful result is not a polished response. It is an order fulfilled accurately and on time.
This is where a discussion apparently about technology becomes a discussion about management. Who maintains the records? Which version is authoritative? Who can approve an exception? Who takes responsibility when an automated recommendation proves wrong? These questions cannot be delegated entirely to a software supplier. They concern how an organisation distributes authority and holds people accountable.
The underlying problem is not uniquely Bangladeshi. Economists Erik Brynjolfsson, Daniel Rock and Chad Syverson have argued that general-purpose technologies such as AI require complementary investment in processes, business models, and human capabilities. Their research helps explain why purchasing technology and realising productivity gains are not simultaneous events. Organisations must also invest in the less visible changes that make the technology useful.
For Bangladesh, this means the skills debate should extend beyond producing more programmers or teaching employees to write prompts. Those efforts matter, but so does preparing managers to assess an AI system’s limitations and frontline workers to challenge its output. An accounts officer who notices an unexplained discrepancy is not an obstacle to automation. That person’s judgement should help determine where automation is appropriate. Training that teaches employees to generate answers without teaching them to question those answers would leave a crucial part of the work unfinished.
Local language adds another dimension. UNESCO’s 2025 AI Readiness Assessment for Bangladesh recommends developing high-quality datasets in Bangla and minority languages, including attention to local meanings and usage. This is an important qualification to any claim that sufficiently capable technology is already available for every purpose. For a public-facing service, fluent Bangla should be the beginning of the test, not its conclusion. Can the system interpret a citizen’s question correctly? Does its answer reflect the actual procedure? Is there a person to approach when the answer is wrong? A citizen should not have to understand the technology to obtain a service or challenge a mistake.
There are local examples of AI being attached to specific operational problems. In June 2025, Ericsson announced that Grameenphone was working with Ericsson and AWS Gen-AI Lab on an AI-assisted approach to migrating its product catalogue from an older system. The proposed process included interpreting existing products, translating requirements, configuring them and testing the results. The announcement described an early proof of concept, not a completed transformation with independently verified financial returns. Its relevance lies in that specificity. There was an identifiable process to improve and work against which the technology could be tested. The distinction between a promising experiment and a demonstrated result should remain equally clear in public discussion.
The larger national question is who will be able to make this transition. The World Bank’s Frontier Firms report warns against relying solely on Bangladesh’s leading businesses to generate the next wave of productive employment. It identifies a substantial divide between highly productive firms and the less productive businesses employing most formal workers and argues for enabling both groups to grow.
AI policy should take that divide seriously. There is a risk that firms able to finance integration, training, and maintenance will pull further ahead, while smaller businesses are offered little beyond subscriptions and motivational workshops. A country could then celebrate conspicuous examples of adoption without achieving much wider improvement. Nor should implementation become an excuse to overlook basic access. UNESCO’s assessment identifies electricity and connectivity barriers, inequalities in digital access and institutional budget constraints. For organisations facing those conditions, readiness cannot be reduced to managerial willingness. Infrastructure, affordability, and technical capability remain essential.
The argument is not that Bangladesh needs less research or fewer technological ambitions. It is that those ambitions should be connected to institutions capable of delivering results.
The February 2026 draft of the National Artificial Intelligence Policy 2026–2030 recognises part of this challenge. It proposes an implementation cell within the ICT Division, bringing together expertise in project management, public administration, digital transformation, and data science. That is a proposal in a draft, not evidence that the arrangement is operating. But its composition acknowledges that AI delivery requires more than technical expertise alone.
The next test should be whether such commitments translate into clear responsibility, sustained funding, and public evaluation. A government-funded AI project should have to show what improved for the people using the service. Faster processing, fewer errors, and effective routes for correction are more meaningful measures than the number of systems launched.
Business support should follow a similarly grounded approach. Smaller firms need help deciding whether a proposed tool solves a genuine problem and whether they can afford to keep it working. Shared testing facilities, practical implementation support, and transparent evidence from earlier projects would offer something more durable than another demonstration. Where conventional software provides better value, choosing it should count as sound judgement rather than technological backwardness.
The Desh–Daewoo experience does not provide a formula for the AI age. It does, however, remind us that technological progress involves learning how to work differently.
Bangladesh should judge its AI ambitions by that standard. The achievement will not be an office that produces more reports or a business that installs another chatbot. It will be an institution that makes better decisions, honours its commitments, and delivers a service people can trust. That is the difference between acquiring intelligence and putting it to work.
The views expressed in this article are solely those of the author
The writer is a Business Analyst, Astha IT
