What Prompt Engineers Actually Do
Prompt engineering has evolved from a casual "ask it the right way" approach to a rigorous, technical discipline. Modern prompt engineers are responsible for designing system prompts that define an AI model's behaviour, tone, and constraints; crafting few-shot examples to improve accuracy; evaluating model outputs for hallucination and bias; and conducting red-team testing to identify failure modes.
Unlike software engineers, prompt engineers don't write code that compiles. Instead, they iterate on natural language instructions, test outputs against rubrics, and measure success through metrics like accuracy, latency, and token efficiency. This is knowledge work that requires deep domain expertise, communication skills, and familiarity with LLM frameworks like LangChain and prompt management platforms.
Core Prompt Engineering Responsibilities
System prompt design: Creating base instructions that define model behaviour, safety guardrails, and output formats.
Few-shot prompt engineering: Building example-driven prompts that teach the model the desired pattern.
Chain-of-thought design: Structuring prompts to force step-by-step reasoning, reducing hallucination.
Evaluation framework development: Creating rubrics and test sets to measure quality before deployment.
Red-team testing: Identifying edge cases, jailbreaks, and failure modes.
Tool integration: Designing prompts that work reliably within RAG, function-calling, or agent frameworks.
Cost optimisation: Optimising token usage without sacrificing quality.
These are not entry-level tasks. A prompt engineer needs to understand LLM behaviour, read academic research, and communicate findings to product and engineering teams.
Why Prompt Engineering Became a Dedicated Role
Two years ago, nobody hired "prompt engineers." Today, every AI-powered SaaS company and enterprise deploying LLMs internally needs them. The explosion of generative AI applications created a gap: software engineers focus on infrastructure and API integration, but someone needs to own the language logic itself.
A badly-tuned prompt wastes tokens, outputs low-quality results, and erodes user trust. A well-tuned prompt reduces support costs, improves customer satisfaction, and becomes a competitive moat. Companies now treat prompt engineering as a specialisation because the output directly impacts revenue and retention.
UK firms report that finding internal prompt engineers is extremely difficult. Academic computer science programmes don't teach it. Most candidates come from hybrid backgrounds—NLP researchers, content strategists, or senior QA engineers retraining themselves. The market is supply-constrained, and salaries are rising.
The UK Talent Supply Problem
According to the UK Government's Office for National Statistics (2024), there are approximately 380,000 software engineers in the UK workforce. Specialist roles like prompt engineering represent less than 2% of this cohort. London-based AI companies report that prompt engineer hires are taking 4–6 months, with salary expectations of £45,000–£60,000 annually for mid-level candidates.
The pipeline from education to job is broken. Universities offer AI and machine learning degrees, but prompt engineering—a nascent discipline—is not yet mainstream curriculum. Most UK prompt engineers are self-trained via online courses or internal reskilling programmes, creating a narrow talent pool competing for high-salary roles.
Geographic concentration is another constraint. Prompt engineering talent clusters in London, Cambridge, and a few tech hubs. Distributed hiring from continental Europe or the US reduces some pressure, but time zones, visa costs, and cultural fit create friction.
Kenya's English-First Graduate Advantage
Kenya has a distinctly different talent dynamic. English is the official language of higher education. Kenya's university system (University of Nairobi, Strathmore, JKUAT, and others) produces graduates in computer science, information systems, and mathematics who are fluent in English from day one. There is no translation step, no accent barrier, no cultural assumption gap.
Kenya's tech talent pool has grown 12% year-on-year (World Bank ICT Workforce Survey, 2023). Nairobi is home to over 300 active tech startups, and major companies like Google, Microsoft, and IBM operate research and engineering centres there. This creates a culture of AI and machine learning work, access to knowledge, and peer learning.
For prompt engineering specifically, Kenya offers:
English-first education: No language barrier; candidates communicate with UK product and engineering teams naturally.
AI research ecosystem: Proximity to academic and industry AI work; candidates stay current with latest LLM research.
Salary efficiency: Mid-level prompt engineers in Nairobi earn £12,000–£15,000 annually, reducing total team cost by 75% vs UK.
Time zone advantage: UTC+3 overlaps with UK working hours (8 hours difference), enabling synchronous collaboration.
Stability and growth: Kenya's tech talent market is expanding, not shrinking; retention is strong for well-managed teams.
Team Structure and Roles
A dedicated prompt engineering team requires a mix of seniority levels and specialisations. Here's a typical structure for a UK company outsourcing to Kenya:
Comparison
| Role | Responsibility | Typical Cost (Kenya) |
|---|---|---|
| Prompt Engineering Lead (1) | Owns strategy, LLM selection, evaluation frameworks, quality standards. 5+ years experience with LLMs or NLP. | £18,000–£22,000 |
| Senior Prompt Engineer (2) | Design system prompts, few-shot engineering, red teaming. Mentors junior team. | £14,000–£18,000 |
| Prompt Engineer (3–4) | Execute daily prompt iteration, evaluation, documentation. Support specific product areas. | £10,000–£13,000 |
| QA/Evaluator (2) | Run rubric-based testing, flag quality issues, create test datasets. | £8,000–£10,000 |
This 8–10-person team, led by a UK-based product manager or engineering lead, costs approximately £110,000–£150,000 annually. The same team in London would cost £500,000+.
Tools and Platforms
A modern prompt engineering team uses both open-source and commercial tools:
Prompt Management and Testing
- LangChain (open-source): Framework for building LLM applications with modular prompts, chains, and memory.
- Prompt.io: Lightweight prompt versioning and testing platform; integrates with OpenAI and other APIs.
- Humanloop: Collaborative prompt engineering with A/B testing and feedback loops.
- Weights & Biases: Experiment tracking for prompt variations and output quality metrics.
Evaluation and Red Teaming
- OpenAI's evals: Framework for evaluating LLM outputs against custom rubrics.
- Arize AI: Production monitoring for LLM outputs; detects drift and quality degradation.
- Scale AI: Human-in-the-loop evaluation service for large-scale rubric testing.
- A well-resourced team uses 3–5 tools concurrently, with clear ownership and integration into CI/CD pipelines.
Cost Comparison: UK vs Kenya
Comparison
| Line Item | UK (London) | Treba (Nairobi) | Saving |
|---|---|---|---|
| Prompt Engineering Lead | £50,000–£60,000 | £18,000–£22,000 | 62% saving |
| Senior Prompt Engineer (2) | £80,000–£100,000 | £28,000–£36,000 | 65% saving |
| Prompt Engineer (3–4) | £120,000–£160,000 | £40,000–£52,000 | 68% saving |
| QA/Evaluator (2) | £60,000–£80,000 | £16,000–£20,000 | 73% saving |
| Total (8–10 FTE) | £410,000–£500,000 | £102,000–£130,000 | 75% saving |
Key takeaways
• Prompt engineering is now a distinct discipline, not a peripheral skill. • UK talent is scarce, expensive, and slow to hire; expect 4–6 month recruitment cycles. • Kenya's English-first education system, AI ecosystem, and lower costs create a strategic advantage. • A dedicated team of 8–10 prompt engineers costs 75% less in Kenya than in the UK. • Successful offshore prompt teams require clear product strategy, documented standards, and weekly sync with UK leadership.
Written by
Treba Research
Treba editorial team — expert analysis on outsourcing, compliance, and building distributed UK–Kenya teams.
