On 23 August 2026, an AI conference in Dhaka gathered corporate leaders, policymakers and technologists to ask how artificial intelligence will reshape industry and employment. That same night, around midnight, a sewing operator — called Rahima in Prothom Alo’s account, a pseudonym — was heading home from a factory in Gazipur, with no idea that her workplace was the subject under discussion.
Where AI is actually arriving in the sector
Bangladeshi ready-made garment manufacturers are weighing AI for production planning, automated quality testing, early warning of machine breakdowns and demand forecasting. None of that is speculative; all four are products a factory can buy today.
The government has released a draft National AI Policy for 2026–2030 for public comment, with final approval still pending.
Which jobs are exposed, and why
Repetitive, monotonous work is the most exposed, for the plain reason that machines do it cheaply and well. If those tasks are automated and skill-based work is not upgraded through training, workers fall behind — and the training has to reach both the factory’s management and its floor, because a plant that buys a quality-testing system and does not teach anyone to operate it has bought an expensive ornament.
The piece makes the upgrade argument concretely: a sewing operator who learns to run quality-testing machines becomes more valuable, not less. That is the whole policy question in one sentence.
The conference line, and what it leaves out
A point raised in Dhaka was that Bangladesh does not need to build the world’s largest AI model; it needs to apply AI effectively to its own problems. That is correct, and it is the right strategic instinct for a country that cannot win a compute race.
It is also incomplete, because applying AI to your own problems is not free either. It requires people who can specify, procure, integrate and maintain these systems — and a factory that cannot assess a vendor’s claims will buy whatever is demonstrated most impressively. The gap is not model-building talent. It is the far more ordinary capacity to be a competent buyer.
What the policy would have to contain to matter
Three things, none of them glamorous. A funded retraining commitment tied to automation — if a factory installs a system that displaces a role, the training to move those workers should be part of the investment, not a hope. A procurement standard, so that a mid-sized factory in Gazipur has a checklist for judging an AI vendor. And a public dataset policy, since the industrial data that would make Bangladesh-specific models useful currently sits unshared inside individual firms.
Rahima will not read the National AI Policy. Whether it is written well is still the thing that decides what her job looks like in five years, and the consultation period is the only point at which anyone outside the room can affect it.
One number frames the stakes. The ready-made garment sector employs roughly four million people in Bangladesh, the majority of them women, and it accounts for the overwhelming share of export earnings. A technology that alters even a modest fraction of those roles is not an industrial story here. It is a macroeconomic one, and it arrives through procurement decisions taken factory by factory, with no central moment at which anyone chooses.
Related: why AI still handles Bangla badly.
