• I used an LLM to help draft this post, but I’ve edited/rewritten it extensively and endorse it. All data is originally researched and verified by me.

@EU AI Governance Effective Charity ≠ Effective Development (Global Health Edition) 

 A Framework for Thinking about the EA Labor Market 

This essay examines AI-driven labour displacement in Kenya as a specific governance risk pathway. The argument is that labour market disruption from automation does not stay contained as an economic problem — it erodes the legitimacy of the institutions that carry the AI governance mandate. When 86% of a workforce is informal and outside the reach of labour law, retraining programmes, and social protection, AI displacement becomes invisible to those institutions. Institutions that cannot see displacement cannot respond to it. That is the pathway I map.

I am an independent researcher based in Nairobi. This is original research built on ILO and World Bank data on Kenyan labour markets. I am pursuing AI governance fellowship applications and publish here to reach people working on AI governance in the Global South. Comments and pushback welcome.

Introduction

Seventeen million Kenyans are in employment (ILO, 2022). Three-quarters of them sit at lower or upper secondary education. Those two numbers carry the weight of this essay.

Kenya sits on the frontier of two trends at once. The country has built a real BPO and digital services economy over the past decade. It has also just published a National AI Strategy that names talent and inclusion as cross-cutting priorities. These facts are related. Both depend on the same working population. Both assume that the people already in those jobs can absorb what comes next.

AI-driven automation is testing that assumption faster than the policy architecture can respond. This essay argues that labour market disruption in Kenya is a pathway by which AI itself becomes harder to govern, and that treating it as a welfare problem alone misreads what it does to the institutions that are meant to steer frontier systems. When a large share of a country's working population loses bargaining power, institutions lose legitimacy, and those same institutions carry the AI governance mandate.

The question I return to through the essay is this: how could AI-driven automation create systemic risk in African economies through labour market disruption, and what could governance do to reduce it?

Section 1: The Labour Market AI Is Entering

Kenya's labour market is unusual and poorly understood from outside the country. Of the 17 million workers ILO counts, 7.28 million sit at upper secondary level, which is 43% of all employment. These are the workers ranked highest for AI exposure in almost every task-displacement study of the last three years. They do clerical work, call-centre work, routine data processing, content moderation, and mid-skill services. This is the band most concentrated around Nairobi and the urban BPO corridors.

A further 12.6 million workers sit at lower or upper secondary combined, which is 74% of the workforce. That is the denominator the National AI Strategy has to work with. The talent pipeline for AI governance assumes skilled graduates. The population actually exposed to AI deployment is a different cohort entirely.

Two structural features make this more serious than a headline count suggests. Young workers aged 15 to 24 are the most concentrated in intermediate and upper secondary roles, with close to 60% in that band. Women are overrepresented in BPO-adjacent jobs and hold roughly half the advanced credentials men hold on average. The workers with the least educational buffer also have the least political weight when displacement hits.

Section 2: The Structural Argument for Displacement Risk

The most important number in this analysis is not 17 million. It is 86.

Eighty-six percent of Kenya's workforce is informal. No contracts. No NSSF coverage. No union. No grievance channel. When a worker in this group loses income to automation, no institution registers the loss. This is not a welfare problem waiting to be addressed. It is an invisibility problem that makes governance structurally impossible.

Set that aside for a moment and look at where the displacement risk sits. The 7.5 million workers in the Intermediate band (upper secondary education) carry most of it. Task-displacement studies rank this cohort consistently as the highest-risk group for AI substitution. They do the work that AI systems are currently best at taking: clerical tasks, call-centre scripts, routine data processing, content moderation, mid-skill services. Kenya's BPO corridor is built on exactly this cohort. So is the retail sector, land transport, and food and beverage services.

The concentration across those three sectors matters. Retail trade employs roughly three million workers at a 96% informality rate. Land transport — bodaboda, matatu — is 91% informal. Food and beverage follows at 87%. These are not abstract risk categories. They are the specific sites where e-commerce, ride-hailing platforms, and AI-driven logistics are already restructuring work, not by eliminating it formally, but by degrading its terms while keeping the worker nominally employed.

Education does not solve this. It reduces it. Informality falls from roughly 70% at below-basic education level down to 26% at Advanced. But that still means one in four university graduates works informally in Kenya. For the Intermediate band, the estimated informality rate sits around 46 to 50 percent. Roughly half the workers most exposed to AI displacement have no formal employment structure beneath them. The BPO narrative assumes contracts. The data says that assumption is wrong for the majority.

The earnings picture closes the argument. Intermediate workers earn approximately 40,000 KES per month less than Advanced workers. Inside that average, the median is lower — a small number of high earners pull it up, so the typical Intermediate worker earns less than the typical statistic implies. There is no savings buffer. No severance. No financial margin to carry a worker through the period between displacement and re-entry. For this group, an AI shock is not a career disruption. It is a household crisis.

Section 3: The Systemic Risk Pathway

The governance failure is not that Kenya lacks an AI strategy. It published one in March 2025. The failure is that the strategy was designed for institutions that can reach 14% of the people who need them.

Labour law applies to formal employment. Retraining programmes require formal employment relationships to activate. Social protection requires registration that most informal workers do not have. The mechanisms the Kenya National AI Strategy 2025–2030 can deploy (cross-cutting priorities, talent pipelines, inclusive AI) are built on a formal economy that covers one in seven Kenyan workers.

This is the systemic risk. Not just displacement, but displacement in a context where governance cannot see it. When an informal retail worker in Nairobi loses income because an e-commerce platform cuts small trader margins, no data system captures that as an AI impact. No institution flags it as a governance failure. It appears in the numbers as employment, because the worker is still technically working.

Two dimensions compound the risk further. Young workers aged 15 to 24 are the most concentrated in the Intermediate band, at 60% compared to the workforce average of 43.7%. Women are overrepresented in BPO-adjacent roles and underrepresented in Advanced credentials. Female youth carry the highest unemployment rate of any cohort in the data. A gender-neutral governance response will protect men more than women by design. The costs fall heaviest on the group with the least institutional weight to demand otherwise.

The NEET rate — nearly one in three young Kenyans not in employment, education, or training — describes the talent pipeline the Strategy assumes exists. It does not exist at the scale the document implies. The workers already in the system are more precarious than the official unemployment rate suggests. The workers entering it face a market that is already saturated.

None of this is an argument against deploying AI. It is an argument about sequencing. Governance systems that cannot reach informal workers, that have no mechanism to count informal displacement, and that have not built the social protection floor that would catch displaced workers — those systems will lose legitimacy faster than AI can generate the productivity gains that justify the disruption. That is the pathway from labour market shock to governance crisis. Kenya is not at the end of that path. But it is walking it.

Section 4: What Good Governance Could Look Like

Kenya already knows how to count people at scale. NEMIS tracks every student in the public education system — enrolment, progression, dropout. SHIF requires every citizen to register for health coverage regardless of employment status. Both systems reach beyond formal employment. Neither principle has been applied to labour. That is the gap.

The first thing good governance does is make the informal economy visible to the state. AI displacement in Kenya will not look like mass redundancies at a Nairobi tech firm. It will look like a slow drop in M-PESA receipts for bodaboda drivers, falling order volumes for River Road printers, declining task rates for freelance data workers. None of these appear in any government data system as an AI governance event. They appear, if at all, as demand fluctuations. The Kenya National Bureau of Statistics runs periodic household surveys on a five-year cycle. That cycle is too slow to catch displacement that arrives through a platform algorithm update. A live labour dashboard, built from M-PESA merchant category data and ODPC-regulated platform disclosures, would give the state a real-time signal before the damage becomes invisible. No new institution is required. Kenya already has what it needs — the mandate just needs extending.

The second step is sector-specific risk registries for retail, land transport, and food and beverage. These three sectors employ millions at 87 to 96 percent informality, and they are the same sectors where AI-driven logistics, ride-hailing algorithms, and e-commerce platforms are already restructuring work. NTSA has the power to revoke operating licences for ride-hailing platforms. Extending that power to require disclosure of how platform algorithms affect worker earnings — verified against M-PESA data — creates a transparency obligation without a new regulatory body. Start with transport, where the institutional teeth already exist. Extend to retail and food and beverage as the data infrastructure matures.

The third requirement is portable benefits. Togo's Novissi programme reached 25 percent of the adult population through mobile phones within weeks during the COVID-19 crisis, with no employment contract required. The technology already exists in Kenya. M-PESA is the delivery mechanism. The gap is the policy decision to link NSSF contribution eligibility to wallet activity rather than a payroll office. That decision has political obstacles — NSSF and the formal employment lobby have no incentive to extend coverage to workers who contribute irregularly. Naming that obstacle honestly is part of what good governance requires.

None of these interventions is gender-neutral by design. The AU Continental AI Strategy names gender as a structural risk but proposes no mechanism to reach women specifically. A response that does not correct for the overrepresentation of women in AI-exposed sectors and their underrepresentation in advanced credentials will reproduce the inequality it was meant to address. The data infrastructure, the registries, and the benefit triggers all need gender disaggregation built in from the start — not added as a correction after the fact.

Conclusion

The numbers in this essay are not projections. Seven and a half million Kenyan workers sit in the education band most exposed to AI task substitution, right now. Eighty-six percent of the workforce is invisible to the governance systems that are meant to protect them. The 40,000 KES earnings gap between Intermediate and Advanced workers means there is no financial margin between displacement and crisis.

The governance problem is not that Kenya lacks ambition. The National AI Strategy has real priorities. The AU Continental AI Strategy explicitly calls for African-led research on labour market risks. The ILO has framework language for workers left behind by structural change. What none of them has is a mechanism that reaches the 86 percent.

I started this research watching work disappear — platforms collapsing, freelancers losing income to automation, skills made obsolete before they could pay back what they cost to learn. The data confirmed what I was seeing. The governance gap confirmed why nobody was counting it.

That is what has to change first. Before retraining. Before policy frameworks. Before AI governance can claim to protect Kenyan workers, it has to be able to see them.

References

Primary Data

International Labour Organization (2022). Employment by sex, age and level of education — Kenya. ILOSTAT database. Indicator: EMP_TEMP_SEX_AGE_EDU_NB_A. Geneva: ILO. https://ilostat.ilo.org/

International Labour Organization (2022). Informal employment by economic activity (ISIC-Rev.2) — Kenya. ILOSTAT database. Indicator: EMP_NIFL_SEX_EC2_NB_A. Geneva: ILO. https://ilostat.ilo.org/

International Labour Organization (2022). Mean nominal monthly earnings of employees by sex and education — Kenya. ILOSTAT database. Indicators: EAR_EMTM_SEX_EDU_NB_A; EAR_EMTA_SEX_EDU_NB_A. Geneva: ILO. https://ilostat.ilo.org/

International Labour Organization (2022). SDG Indicator 8.3.1 — Proportion of informal employment in non-agriculture employment, by sex — Kenya. ILOSTAT database. Geneva: ILO. https://ilostat.ilo.org/

International Labour Organization (2022). SDG Indicator 8.6.1 — Share of youth not in employment, education or training (NEET rate) — Kenya. ILOSTAT database. Geneva: ILO. https://ilostat.ilo.org/

World Bank (2024). World Development Indicators — Kenya: youth unemployment, female labour force participation, secondary school enrolment. Washington DC: World Bank. https://databank.worldbank.org/source/world-development-indicators

Policy Documents

Kenya Ministry of ICT and Digital Economy (2025). Kenya National Artificial Intelligence Strategy 2025–2030. Nairobi: Government of Kenya.

African Union Commission (2024). Continental Artificial Intelligence Strategy: Harnessing AI for Africa's Development and Prosperity. Addis Ababa: African Union Commission. July 2024.

International Labour Organization (2015). Guidelines for a just transition towards environmentally sustainable economies and societies for all. Geneva: ILO. https://www.ilo.org/wcmsp5/groups/public/---ed_emp/---emp_ent/documents/publication/wcms_432859.pdf

Republic of Rwanda, Ministry of ICT and Innovation (2022). National Digital Inclusion Strategy. Kigali: Government of Rwanda.

Research and Secondary Sources

Frey, C.B. & Osborne, M.A. (2017). The future of employment: How susceptible are jobs to computerisation? Technological Forecasting and Social Change, 114, 254–280. https://doi.org/10.1016/j.techfore.2016.08.019

Acemoglu, D. & Restrepo, P. (2022). Tasks, automation, and the rise in US wage inequality. Econometrica, 90(5), 1973–2016.

GSMA Intelligence (2021). The Novissi programme: Mobile-enabled social protection in Togo. London: GSMA. https://www.gsma.com/mobilefordevelopment/

AI Safety Atlas (2024). Systemic Risks. AI Safety Atlas, Chapter V1. https://ai-safety-atlas.com/chapters/v1/risks/systemic-risks/

Crossposted from my LinkedIn newsletter, A Couple of Thoughts.