For years, one route into Bangladesh's digital economy looked fairly clear. Learn basic coding, Photoshop, or WordPress. Find small jobs from overseas clients. Build experience through background removal, template websites, content writing, data work, or junior support. The work was rarely glamorous, but it gave people a place to begin.
AI is making that first step harder to find.
Around the world, the debate over AI and employment swings between optimism and alarm. The World Economic Forum projects that millions of new roles will emerge even as others disappear. The IMF estimates that roughly 40 percent of jobs globally are exposed to AI in some capacity. The International Labour Organization argues that generative AI is more likely to transform tasks than completely replace occupations.
The headline numbers do not tell us what happens at the entrance to a career. AI can compress routine cognitive work without eliminating an occupation. Bangladesh's digital economy happens to rely heavily on the kind of entry-level tasks that are easiest to compress.
In advanced economies, companies are not dismantling entire departments. They are adjusting hiring patterns. A team that once required five junior analysts may now need two. A marketing firm that once hired three entry-level content writers may now hire one editor who works alongside AI tools. Startups that previously recruited several junior developers for boilerplate coding can now generate and debug significant portions of code through large language models.
The result may be fewer junior openings, even when a company is still growing.
We sometimes see the headlines. In 2024 and 2025 alone, major global technology firms such as Google, Amazon, Meta, Microsoft, and Salesforce announced tens of thousands of layoffs, with the broader tech sector cutting well over 200,000 jobs worldwide. In India, several IT services giants slowed hiring sharply after pandemic-era expansion, while startup layoffs made regular news. Not all of these cuts were purely AI-driven; some were corrections after over-hiring. But they signal something: companies are recalibrating workforce size in an era where automation and AI tools allow them to do more with fewer people.
Research from MIT Sloan offers a clearer picture of this productivity shift. Studies show that generative AI increases software developers' task output by around 26 percent. Less experienced workers saw even larger gains, between 27 and 39 percent. In writing and coding tasks, AI boosts efficiency when used as a complement to human labour rather than a replacement.
But productivity gains carry labour-market consequences. If output per engineer rises by roughly a quarter, what ten engineers once produced can now be delivered by seven. That does not automatically mean layoffs. It means slower hiring, smaller junior intakes, and leaner teams.
That is different from mass unemployment. It is a problem with apprenticeship: if the routine work disappears, beginners lose the work through which they once became experienced.
Bangladesh enters this transition from a different economic position than Silicon Valley or Berlin. Our labour market remains highly informal. Manufacturing and services dominate. Many physical jobs remain difficult to automate. But our vulnerability lies elsewhere.
Graduate unemployment in Bangladesh has consistently been higher than that of less-educated groups. Labour Force Survey data from recent years show that tertiary-educated youth face double-digit unemployment rates, significantly above the national average. That structural fragility existed even before AI accelerated.
At the same time, Bangladesh's IT and freelancing growth story was built on scalable digital micro-tasks: background removal for e-commerce images, template-based websites, basic coding products, digital marketing content, data formatting and cleaning. These tasks were globally tradable and required moderate skill. They offered entry into the digital economy.
AI tools now handle parts of this work quickly. Product photos can be edited automatically. Website templates and marketing copy can be generated, and code can be drafted or refactored with assistance.
None of these sectors will disappear overnight. But they will require fewer humans to produce the same output. That is the compression effect.
Bangladesh has lived through earlier waves of mechanisation and digitisation in garments, banking, and telecom. This wave reaches a different category of work: the entry-level cognitive tasks that many graduates expected to use as a starting point.
For a developing economy trying to move workers upward from low-skill labour into digital services, that shift matters a lot. If companies increasingly demand mid-level, AI-supervising talent instead of entry-level operators, the pipeline problem becomes urgent. Without junior roles, how does the next generation accumulate experience?
Covid accelerated the move of classrooms, meetings, and services onto screens. That made more work available to software and, later, to automation. AI did not create the transition, but it has made the consequences harder to ignore.
Meanwhile, universities continue producing graduates at scale. Degrees expand. Aspirations expand. But the structure of tasks available to absorb those graduates is changing faster than curriculum reform.
From a macroeconomic perspective, AI may boost productivity and export competitiveness. Firms can scale output with fewer workers. Service providers can operate more efficiently. But productivity growth does not automatically translate into employment growth. If output per worker rises substantially, firms may expand revenue without proportionally expanding payroll.
This is where inequality risks surface. A smaller group of highly AI-proficient workers may see rising incomes, while entry-level graduates face wage stagnation or underemployment. For a country that frequently invokes the "demographic dividend," distribution matters. Youth is only an asset if it can be productively absorbed.
Another challenge is measurement. Bangladesh lacks granular occupational exposure mapping to AI. We do not have detailed, task-level public data showing which professions are most vulnerable. Disruption may appear socially before it appears statistically. Freelancers may see shrinking contracts before official reports register a change. Graduate frustration may rise before unemployment rates spike.
By the time data confirms the shift, adaptation may already be too late.
Bangladesh's digital strategy has focused heavily on expanding access, expanding connectivity, promoting freelancing, and digitizing services. Those achievements are significant. But access alone does not ensure resilience. If our digital economy depends heavily on low-complexity cognitive exports, AI's rapid improvement directly challenges that comparative advantage.
The safer skills are moving toward designing systems, judging AI output, and combining technical ability with knowledge of a particular domain. Telling every student to "learn AI" is too vague. They need practice defining problems, checking results, communicating decisions, and working beyond a template.
AI is not inherently a threat. As MIT research suggests, it works best as a complement to human labour. Used wisely, it can raise productivity, improve public services, and expand innovation. But complements can still shrink teams.
If ten engineers can now do the work that once required thirteen, firms will adjust accordingly.
Bangladesh is unlikely to see dramatic AI-driven layoffs dominating headlines. The risk is quieter: slower hiring, higher skill thresholds, rising competition for fewer junior roles, wage pressure at the lower end, graduate unemployment creeping upward.
The ladder will not vanish overnight. It may simply admit fewer people at the bottom.
Bangladesh does not need another broad promise about a digital future. It needs better evidence about which tasks are changing, curricula that respond to that evidence, and routes for graduates to gain real experience. Otherwise, we will keep preparing young people for the jobs through which the previous generation entered the industry.