Pakistan’s AI skills debate is getting more practical. That is welcome. But the country still needs a better answer to a basic question: what does it mean for an ordinary worker to be genuinely AI-enabled?
At the Pakistan Women Leadership and Policy Summit in Islamabad on September 10, the message was that AI training should extend far beyond programmers. Pakistan needs AI-enabled plumbers, electricians, technicians, farmers, teachers, nurses and entrepreneurs. That is the right direction because most economic value from AI will come from people who already understand an occupation and learn to use the technology inside it.
The risk is that a broad training push becomes a broad certificate push.
Pakistan has already set ambitious national targets. In February, Prime Minister Shehbaz Sharif announced plans to train one million non-IT professionals in AI skills by 2030. If that effort succeeds, the country will have a much larger pool of people who have encountered AI in formal training. But exposure and capability are different things.
A worker becomes valuable with AI when the technology improves the final result without weakening judgment, accountability or reliability. That is difficult to measure with a course-completion certificate.
A better model is an occupation-specific AI capability ladder.
The first level is assisted use. The worker can use an approved AI tool to produce a draft, analyze information or generate options for a familiar task. A teacher might create a first draft of a lesson plan. A salesperson might prepare outreach. A technician might summarize a maintenance record.
The second level is verification. The worker can identify likely errors, check important claims, compare the output against reliable sources and correct weaknesses before the work reaches a customer, manager or citizen. This is where many productivity claims become less impressive. A draft produced in two minutes does not save time if someone spends 20 minutes repairing it.
The third level is exception handling. The worker knows when the AI is outside its competence, when a situation is unusual, and when human review must take priority. This matters in every occupation. A tool that works on routine cases can still fail badly when the facts are ambiguous, the data is incomplete or the consequences are high.
The fourth level is outcome ownership. The worker can use AI while remaining responsible for the final result. At this point, the person is no longer simply operating a tool. The person is redesigning work around it while preserving professional judgment.
Training programs should be built around this ladder. Instead of asking only whether someone attended a course or passed a quiz, providers should require a representative work sample at each level. Employers could then evaluate what a certificate actually means.
The labor-market benefit would be substantial. Workers worried that AI threatens their jobs often see training as participation in their own replacement. A capability ladder changes that calculation. It makes AI competence a visible career asset. An employee who can verify AI output, handle exceptions and own the result becomes more useful inside the current organization and more employable outside it.
Employers also gain a better way to make decisions. Rather than declaring that everyone must “learn AI,” they can specify what each role actually requires. A customer-service representative may need assisted use and verification. A finance manager may also need exception handling. A senior professional responsible for regulated or high-stakes decisions may need full outcome ownership.
Pakistan is already seeing examples of learning tied more closely to work. JazzWorld’s new AI Associates Program puts 100 young professionals alongside existing teams to solve real business and customer problems rather than isolating them in a separate AI unit. That kind of practice matters because people learn the limits of AI only when they use it in situations where the answer has consequences.
The country should build on that principle as national training expands. Measure whether workers can perform representative tasks before training, after training and again 30 days later. Count final quality, total time including verification and correction, and the ability to recognize cases that require human judgment.
Pakistan does need more AI-enabled workers. But the number of certificates issued will tell us very little about whether the country has produced them.
The better measure is whether people can use AI to do real work well, catch it when it fails and remain accountable for the outcome.
BY: Writer Gleb Tsipursky, PhD, a behavioral scientist, CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work
