Thought Behind Things
Why more AI means more demand for human skills
Dr. Sarah Khan on measuring soft skills for Pakistan and the Global South, and why AI raises the value of the skills it cannot replace.
Contents
- Why Muzamil wanted this conversation
- From one-on-one counseling to skills at scale
- Measuring without flattening people
- From 180 behaviors to 10 core skills
- How the assessment works, and where AI helps
- A baseline instead of a one-day workshop
- Muzamil’s pitch: make it an API
- What AI actually does to jobs
- The 90-day sprint
- Domain experts plus AI
Why Muzamil wanted this conversation
The episode opens with Muzamil explaining why this guest matters to him. He says he has spent a long time looking for women leaders to feature, because younger girls need role models they can watch and learn from. The second reason is the topic itself. About a year and a half earlier he travelled through the US, visiting New York, San Francisco, Seattle and Dallas, and he came away with one observation: Pakistanis are not short on effort, but they lack soft skills and the exposure that global jobs require. His question is whether the mechanisms are missing, and whether fixing them would let Pakistani talent do better work.
From one-on-one counseling to skills at scale
Dr. Sarah Khan traces her path back nine to ten years, to a master’s program in psychology. She found that counseling is reactive: you wait for a person to arrive with a problem, and you serve one person at a time. That is not scalable. She wanted work that reached many people at once, and that led her to social-emotional learning. It became her master’s and PhD, and then work with a global entity, including a research lab at Harvard University.
Over time the work became sector agnostic. Wherever there are humans, Sarah Khan says, these skills are relevant, whether in education, manufacturing or banking. She names the central tension early: “it’s quite a paradox,” because the more AI comes in, the more the importance of human skills grows. The old questions remain. What are soft skills, how do you define them, and above all how do you measure them?
Measuring without flattening people
Her measurement problem has two requirements. The instrument has to be scientifically valid and reliable, so the results can be trusted. But too much standardization reduces a person to a fixed category and loses subjectivity and cultural nuance. She says she faced this as a researcher and practitioner from the Global South.
The starting models did not fit. Sarah Khan describes how Western frameworks, such as emotional intelligence models, describe resilience and collaboration very differently from how those skills show up in practice among her people. If the theory does not fit, the practices that follow from it cannot either. So over about nine years her team built its own theory, which she calls bifocal agency. Thrive by Design is the client-facing version, and from it come the practices, starting with an assessment tool.
From 180 behaviors to 10 core skills
Sarah Khan explains how the field got organized. A Harvard lab mapped the world’s soft skill frameworks, which were numerous and often described the same thing under different names. One framework says resilience, another says grit. The lab sorted them into a master list of 6 categories, 23 subcategories and 180 behaviors. Each behavior is either happening or not, which makes it simple to observe.
That master list worked as what she calls a universal adaptor. A team could take the World Bank’s framework or Singapore’s 21st century framework and map it onto the same behaviors. From there her team condensed the list into 10 top skills that they consider most relevant to the workforce in the Global South.
How the assessment works, and where AI helps
The Thrive by Design assessment is digital. It can be self-reported, or run as a three-way assessment where people also rate peers and their supervisor. That depends on how committed the organization is, since it takes alignment and time.
The assessment has closed-ended items and open-ended ones. Closed-ended scores come back immediately, broken down by skill layer. Open-ended responses are kept because the construct is qualitative and cannot simply be reduced to numbers. This is where AI comes in. Sarah Khan says a task that used to take about two weeks now takes twenty minutes, with accuracy still around 98 percent. Each person gets an individual report, not a generic one, with their strengths and areas to improve. Leadership gets an aggregate view.
She also describes a layered model. There is a foundational layer of skills, and if it is not in place, the chances of building the next layers are lower.
A baseline instead of a one-day workshop
She compares assessment to a doctor’s visit. When something feels wrong, the doctor first runs checks before prescribing anything. The same logic gives a company a number to start from.
This is the distinguishing factor she names: “We are very much focused on quantifying the ROI.” She contrasts it with one-day workshops, where she says companies end up investing around 5 to 6 million. Her ideal client is one that has already spent years building a sophisticated learning and development structure and still could not measure the impact. She says the ideal client is the one who finds her. Those clients arrive understanding the problem and pay well for the time and investment.
Muzamil’s pitch: make it an API
Muzamil pushes on a different use case. Beyond large organizations, he wants a tool that a community or education business could plug into. He imagines a pay-as-you-grow model: a third party sends people to take the assessment, the logic stays with Thrive by Design, and the client gets back a number or a report and pays per call. He admits he was looking for exactly this himself: “I was looking for a solution like this, to be honest.”
Sarah Khan agrees it makes sense and says some organizations are asking for something similar. Their concern is that they do not want data to leave their walls, so they want a native solution embedded into their existing systems. She describes two intelligence layers her team is working toward. One gives organizations their own embedded intelligence model, built on locked scoring logic, so it is not a random LLM answering. She says the second is more appealing but harder to execute, because it involves norms, cybersecurity measures and data questions. She told Muzamil she had written his suggestion down.
What AI actually does to jobs
Sarah Khan spends a stretch on the fear of job loss. She cites a Harvard Business Review view that people who use AI will replace people who do not. She then cites a World Bank report from a few weeks earlier saying that in developing countries only 4 percent of jobs will be impacted by AI. She also cites the World Bank group president’s figure that over the next 10 to 15 years, 1.2 billion young people will enter the job market while around 400 million jobs are expected. Her conclusion: “it’s not that AI is eating jobs, it’s that not enough jobs are being created.”
On skills, she cites a World Economic Forum report saying 39 percent of job skills will be replaced, with soft skills and creative thinking rising. With three young people applying for each job, she argues the differentiator will not be technical skills or content knowledge. It will be adaptability, open-mindedness and curiosity. AI is a tool, she says, and its output depends on the person using it. If your reasoning is weak, it will produce worse output at a faster pace. If your thinking is good, it refines and extends it.
Muzamil adds the view from his own AI work. He says that since December of last year there have been very real deployments and productivity gains in the US, and he points to Jevons paradox: as costs fall, demand expands and overall jobs can rise. Clients who could never afford this work before may start hiring. But he warns that Pakistani companies often lack mapped workflows, run on trusted intermediaries, and are averse to change. Without repeatable processes, they cannot scale.
The 90-day sprint
When Muzamil asks what happens after an assessment, Sarah Khan says it begins with discovery and a pain point. Examples she gives include a company that cannot fill a position, production units running in overdrive, and sales fraud costing a significant amount every month. The baseline lets her pinpoint where the problem sits.
She says her firm works with six core systems in people management, such as onboarding, recognition and compensation, and learning and development. In a 90-day sprint they pick the system that will help most, agree on the outcome that would count as success, and work on it. She mentions targets such as reducing a monetary loss by 10, 20 or 30 percent, shown with before and after data.
Domain experts plus AI
Muzamil adds a note for viewers. He argues that in Pakistan, tech builders and behavior-change experts can collaborate, because software cannot move an organization unless someone changes the behavior around it.
Sarah Khan ends with a point about who builds products. She relays a comment from a lead engineer that software is now just like texting. From that she concludes that technology is no longer the hard part, and human skills are. Her closing line: “the best person to build an accounting software is an accountant.” Domain experts with real content knowledge, layered with AI and technology, can make a product scalable and repeatable. Muzamil closes by praising her for stepping outside the therapy-chair path to create demand for a new way of working.
Want to be a guest on Thought Behind Things? Apply here: https://muzamilhasan.com/podcast/apply
Build with me in the Inner Circle.
You don't need to code. You need a problem you know well, and a room of people building beside you.

