The Graduate and the Machine

Why we conducted this study

Most research gets done because a client has a budget and a deadline. Some of the most useful questions don't come with a commissioning letter attached. They come from paying attention to what people are actually talking about. By early 2026, AI tools had become a normal part of conversation among Nairobi's young professionals, and a mid-2026 global update on digital behaviour found that 97.5% of Kenyan internet users had used at least one AI tool in the past month, the highest adoption rate of any country tracked.

So we funded the study ourselves. Not because a client asked, but to demonstrate how Sonivance Ltd works: when we see a question worth answering, we don't wait for permission to go and answer it. A modest budget, a tight questionnaire and three weeks of fieldwork were enough to get an honest picture.

The research question

How are young professionals and recent graduates in Nairobi actually using AI tools in their work and job search, and how do they feel about what it means for their careers?

• Which AI tools, if any, are they using day to day?
• What tasks are they doing differently because of AI?
• Is AI a threat, a way to compete, or both?
• Where did they learn, and how confident do they feel?
• What would make them trust AI tools more, or use them less?

Study design

Deliberately narrow, so it could be executed quickly and cheaply while still producing a coherent picture. Respondents were drawn from a mix of Nairobi neighbourhoods, from Kilimani and Westlands through to Kasarani and Umoja, and targeting employed and job-seeking people aged 20 to 30 with at least a diploma. Sampling was purposive and snowball through professional networks, alumni groups and co-working spaces, supplemented by short intercept interviews at two business hubs. This is not a representative sample of Nairobi's workforce, and we're not claiming it is.

Data quality

Fieldwork yielded 200 completed questionnaires. Our supervisor reviewed every submission for completion time, response patterns and duplicate device or contact details. 20 records were dropped as invalid (mostly very short completion times or clearly patterned "straight line" answers), leaving a clean analytical sample of 180 valid respondents.

What we found

The findings below are illustrative of the kind of patterns a study like this surfaces, rather than claims about the national or even city-wide population.

1. Finding 1

   Most respondents use AI tools, but casually rather than systematically

   Around two thirds reported using an AI tool at least weekly (mostly for drafting text, summarising documents or generating ideas), but few had built it into a consistent workflow.

2. Finding 2

   Job seekers use AI more heavily than the employed

   Respondents actively job hunting were considerably more likely to use AI regularly, mainly for CVs, cover letters and interview preparation, treating it as a competitive necessity.

3. Finding 3

   Confidence is uneven and mostly self-taught

   Most learned informally through YouTube, friends or trial and error. Confidence in using the tools well was notably lower than frequency of use would suggest.

4. Finding 4

   Anxiety about AI is real, but specific rather than general

   Concern focused on particular tasks becoming less valuable as standalone skills, among them entry-level writing, basic research and first-draft design.

5. Finding 5

   Many see AI as helping them appear more competent than they feel

   A recurring, candid theme was polishing work quickly under deadline pressure, sometimes without fully understanding what the tool had changed.

6. Finding 6

   Trust depends heavily on the stakes involved

   Comfortable for drafting and idea generation; far more cautious where there are financial, legal or academic consequences.

7. Finding 7

   Few have received any formal guidance from employers

   Most said their workplace had no clear policy or training on AI tool use.

What surprised us

• Job seekers, not employees, are the heaviest users. The sharpest, most deliberate use was among people trying to get hired.
• The anxiety is quieter and more specific. Closer to "this particular skill matters less now" than to panic.
• Nobody is really being trained. The near-total absence of employer guidance stood out.

What the findings mean

For businesses and employers

Young staff are already using AI tools, often without guidance. A short, practical policy setting out what is encouraged, what needs a human check and what is off limits would close real gaps.

For NGOs and development organizations

Youth employability programmes could usefully include AI literacy focused on judgement and verification, not just tool familiarity.

For policymakers

The picture is less about mass job replacement and more about a shifting skills premium, worth addressing task by task rather than as one undifferentiated threat.

For researchers

The gap between frequency of use and confidence in use points to an unaddressed training and quality-assurance need.

For employers hiring entry-level talent

If job seekers already use AI to present themselves, screening built entirely around written application quality may need rethinking.

For young professionals

Real value lies in developing judgement about when to rely on AI output and when to check it, a skill distinct from knowing how to use the tools.

Keeping research affordable

Useful research doesn't have to be expensive to be credible. A short, focused questionnaire; digital data collection from day one; a small, well-briefed field team; targeted rather than random sampling; real-time monitoring on the KoboToolbox dashboard; automated validation and skip logic; lean supervision; in-house analysis; and a short reporting turnaround.

Figures are illustrative only and are not an actual financial statement.

What this demonstrates about Sonivance

Research design → tool development → digital data collection → field supervision → data quality → analysis → insights → reporting. We wrote the questionnaire, built and tested it in KoboToolbox, trained and deployed a small enumerator team, monitored data quality in real time, cleaned and analysed the results, and turned it into a report worth reading, all on a modest, self-funded budget and a short timeline.

Sonivance doesn't need a large national mandate to produce research that's rigorous, honest about its limits, and genuinely useful.

References

• DataReportal, We Are Social & Manochi (2026). Digital 2026: Mid-Year Global Update.
• Kenya National Bureau of Statistics (2026). Kenya Integrated Labour Force Survey.
• KICTANet (2026). Kenya Leads the World in AI Adoption in 2026.
• International Labour Organization (2026). ILOSTAT Labour Force Statistics Database.

External sources are cited for contextual background only. All sample figures, quotations, percentages and findings in this case study are illustrative.