Lessons from Cambridge
Developing organisational AI capability requires more than ad-hoc workshops. This article documents the design and implementation of a comprehensive 24-week AI training programme at Cambridge University Press and Assessment, demonstrating how structured, bite-sized learning can transform busy professionals from “AI curious” to “AI confident” whilst maintaining their daily work commitments.
Introduction
The challenge facing organisations today is not whether to adopt AI, but how to build genuine capability across their workforce. Whilst many professionals use AI tools sporadically, few possess the systematic skills required to integrate AI effectively into their workflows, automate repetitive tasks, or design custom solutions for their specific contexts.
At Cambridge University Press and Assessment, I developed and delivered an internal AI training programme specifically designed for working professionals – individuals with limited technical backgrounds but clear operational needs for AI capability. The programme runs for 24 weeks, with twice-weekly 30-minute sessions, structured to take participants from foundational understanding through to advanced workflow automation.
This article documents the pedagogical approach, curriculum architecture, and practical implementation insights from this programme, offering a replicable framework for organisations seeking to build sustainable AI capability rather than superficial awareness.
Why This Matters: Effective AI training programmes must balance technical depth with accessibility, sustain engagement over months rather than days, and produce tangible outputs that demonstrate capability. The approach detailed here addresses all three requirements whilst respecting the time constraints of working professionals.
The 30-Minute Formula: Programme Design Philosophy
The cornerstone of this training programme is what I term the “30-minute formula” – a structured approach that divides each session into three equal segments:
10 Minutes: Instruction
Core concepts explained clearly, without jargon, grounded in practical business contexts. Rather than theoretical frameworks, instruction focuses on “what you need to know to use this effectively today.”
10 Minutes: Demonstration
Live demonstration of concepts in action. Participants observe real-time AI interaction, seeing both successes and limitations. This transparency builds realistic expectations whilst showcasing practical applications.
10 Minutes: Practice
Immediate hands-on application with provided prompts and scenarios. Participants experiment whilst support is available, building confidence through doing rather than passive observation.
Why This Structure Works:
The 10/10/10 format addresses four critical challenges in professional development:
- Attention span optimisation: Research consistently demonstrates that engagement drops significantly after 20-30 minutes. By containing complete learning cycles within 30 minutes, the programme maintains high engagement throughout.
- Immediate application: Participants practise whilst the instruction remains fresh, reinforcing concepts before cognitive load overwhelms retention.
- Schedule flexibility: Thirty-minute sessions fit into lunch breaks, between meetings, or at the start/end of working days. Longer sessions require calendar space that busy professionals struggle to protect.
- Sustainable commitment: Two 30-minute sessions weekly represents a manageable one-hour weekly time investment over six months—far more achievable than intensive multi-day workshops that disrupt operational work.
This pedagogical approach deliberately prioritises sustainability over intensity. Professional development fails not from lack of initial enthusiasm but from inability to maintain consistent engagement over the timeframes required for genuine skill development.
Curriculum Architecture: Progressive Complexity Over 24 Weeks
The programme comprises five modules, each building systematically on prior learning whilst maintaining practical applicability at every stage.
Module 1: AI Foundations (Weeks 1-4)
Establishing conceptual understanding before technical application:
Lesson 1.1: Introduction to AI and Machine Learning Distinguishing AI from traditional software using accessible analogies. The “calculator versus weather forecaster” comparison proved particularly effective—calculators follow exact rules for precise answers; weather forecasters use patterns to predict likely outcomes. This simple mental model helps participants understand why AI produces probabilistic rather than deterministic results.
Lesson 1.2: Understanding Large Language Models Demystifying how chatbots process and generate text. Covering tokens, context windows, and the phenomenon of “hallucinations”—not to create technical experts but to build informed users who understand tool limitations.
Lesson 1.3: AI Capabilities and Limitations Setting realistic expectations by identifying tasks where AI excels versus struggles. Positioning AI explicitly as an augmentation tool rather than replacement technology reduces anxiety whilst building appropriate use cases.
Lesson 1.4: AI Ethics and Responsible Use Developing ethical guidelines for AI use, covering bias identification, privacy protection, intellectual property respect, and personal ethical frameworks.
Key Innovation: The first practical activity deliberately asks participants to interact with AI in their mother tongue, requesting culturally specific content. This multilingual, multicultural approach immediately demonstrates both AI’s remarkable language capabilities and its occasional cultural blind spots, building nuanced understanding from day one.
Module 2: Mastering Prompt Engineering (Weeks 5-10)
The foundation of effective AI interaction, covering six progressive lessons:
- Fundamentals of prompting (context, clarity, role-playing)
- Zero-shot and few-shot techniques
- Chain-of-thought prompting for complex reasoning
- Advanced techniques (output formatting, conditional logic)
- System prompts and custom instructions
- Building reusable prompt libraries
This extended module reflects a core design principle: time invested in prompt engineering pays dividends across all subsequent AI work. Participants who master prompting can immediately apply these skills in their daily roles, generating quick wins that sustain motivation through later, more technical modules.
Tangible Output: By module completion, each participant creates a personal library of 20 reusable prompts specific to their role—templates they return to repeatedly rather than starting from scratch with each AI interaction.
Module 3: Practical AI Tools and Applications (Weeks 11-16)
Moving beyond text-based interaction to explore AI’s broader capabilities:
- Multimodal AI (working with images)
- Data analysis and visualisation
- Document processing and analysis
- AI-powered writing and editing
- Code generation for non-programmers (Excel formulas, SQL queries)
- Custom GPTs and specialised assistants
This module deliberately emphasises diversity of application. Participants discover which AI capabilities align with their specific work contexts, building personalised tool sets rather than generic skills.
Module 4: Advanced AI Concepts (Weeks 17-20)
Understanding sophisticated implementations without requiring technical expertise:
- Retrieval-Augmented Generation (RAG) systems
- AI agents (planning, tools, memory components)
- Knowledge graphs and structured data
- Model Context Protocol (MCP) for connecting AI to external tools
The focus here is conceptual understanding sufficient for informed decision-making. Participants learn to evaluate whether these advanced approaches suit their use cases and communicate effectively with technical implementers, even if they never build these systems themselves.
Module 5: Workflow Automation and Integration (Weeks 21-24)
Scaling AI impact beyond individual tasks:
- Components of AI workflows (triggers, actions, connections)
- Building simple automations with no-code platforms
- Creating multi-step AI workflow chains
- Integration strategies and security considerations
- Performance monitoring and optimisation
- Continuous learning and community engagement
Tangible Output: Participants design and implement a five-step automated workflow relevant to their actual work, demonstrating capability to move from AI-assisted tasks to AI-powered processes.
Pedagogical Approach: Learning by Doing
Beyond the 30-minute formula and curriculum structure, several pedagogical principles distinguish this programme:
Hands-On Learning Prioritised
Every session includes mandatory practice time. Participants interact with AI tools during sessions rather than listening to descriptions of what AI can do. This approach recognises that AI fluency, like language fluency, develops through use rather than study.
Real-World Applications From Day One
Homework assignments tie directly to participants’ actual work needs: drafting emails they genuinely need to send, explaining complex topics from their fields, planning real meetings or projects. This immediate applicability maintains motivation and demonstrates value continuously rather than promising future benefit.
Scaffolded Support Through NotebookLM
Throughout the programme, curated materials are added to a shared Google NotebookLM collection—an AI-powered knowledge base that participants query between sessions. This provides:
- Curated external resources organised by module
- Audio summaries for those who prefer listening to reading
- Quizzes and flashcards for self-assessment
- Persistent access to an AI assistant trained on course content
NotebookLM extends the learning environment beyond twice-weekly sessions, creating continuous rather than episodic engagement.
Community Learning via Teams Chat
Participants share discoveries, troubleshoot challenges, and celebrate successes through a dedicated Teams channel. This transforms individual learning into collective capability-building, with peer teaching reinforcing concepts whilst building organisational knowledge.
Safe Learning Principles
The programme explicitly establishes psychological safety norms:
- Mistakes are learning opportunities
- No question is too basic
- We support each other’s growth
- Experimentation is encouraged over perfection
These norms prove particularly important for participants anxious about technology or concerned about appearing incompetent. Creating permission to struggle publicly accelerates learning by making visible the common challenges everyone faces privately.
What Makes This Programme Distinctive
Several design choices distinguish this programme from typical AI training approaches:
1. The Calculator Versus Weather Forecaster Analogy
Traditional AI training often begins with technical explanations of neural networks and training data. This programme instead uses an accessible analogy: calculators follow precise rules for exact answers; weather forecasters use patterns to predict likely outcomes.
This simple mental model helps participants understand:
- Why AI sometimes produces unexpected results (it’s predicting, not calculating)
- Why the same input may yield slightly different outputs (probabilistic, not deterministic)
- Why AI requires verification for critical tasks (forecasts need checking)
- Why context matters (weather predictions depend on location, season, historical patterns)
The analogy appears throughout the programme, providing a consistent framework for understanding AI behaviour.
2. Multilingual First Activity
Rather than demonstrating AI with English examples, the first hands-on activity asks participants to:
- Introduce themselves to AI in their mother tongue
- Request a proverb about learning from their culture
- Ask for a tongue twister or local slang explanation
This approach serves multiple purposes:
- Demonstrates AI’s remarkable multilingual capabilities
- Surfaces cultural strengths and limitations
- Reduces anxiety by starting with familiar rather than foreign concepts
- Celebrates linguistic diversity within the organisation
- Creates memorable first experiences that build confidence
Participants from Afrikaans, Arabic, Mandarin, Hindi, Spanish, and numerous other linguistic backgrounds experience immediate success, establishing AI as a tool accessible to them regardless of English fluency.
3. Playful Yet Practical Exercises
Early exercises balance playfulness with practicality:
Exercise 1: “Explain what a mortgage is like I’m 10 years old” versus “Explain what a mortgage is to someone buying their first home”—demonstrating AI’s ability to adjust communication style to audience.
Exercise 2: A complex constraint-satisfaction problem (ordering pizza for 12 people with dietary restrictions and budget limits)—showing AI’s capacity for multi-variable optimisation.
Exercise 3: Creative generation (writing a two-sentence story about an animal working in a specific profession)—revealing AI’s creative capabilities whilst remaining lighthearted.
These exercises deliberately lower stakes whilst teaching transferable skills. Participants discover AI capabilities through engaging scenarios before applying those capabilities to high-stakes work tasks.
4. Explicit Focus on Augmentation, Not Replacement
From the first session, the programme positions AI as an augmentation tool rather than a replacement technology. Key messages reinforced throughout:
- “AI is a tool, you’re in charge”
- “AI makes mistakes—always verify important outputs”
- “AI recognises patterns; it doesn’t ‘think’ like humans”
- “Every expert was once a beginner”
This framing reduces anxiety about job displacement whilst building appropriate expectations about AI’s capabilities and limitations.
Implementation Insights
Running a 24-week training programme revealed several implementation challenges and effective responses:
Challenge 1: Maintaining Participation Over Six Months
The Problem: Initial enthusiasm typically wanes after 6-8 weeks. Professional commitments create schedule conflicts. Participants fall behind and disengage.
Solutions Implemented:
- Session recordings provided for those who miss live sessions
- NotebookLM ensures missed content remains accessible
- Regular check-ins via Teams to maintain community connection
- Milestone celebrations at module completion (e.g., sharing prompt libraries, demonstrating workflows)
- Flexible attendance expectations—participants can dip in and out based on current priorities
Challenge 2: Managing Diverse Technical Backgrounds
The Problem: Participants ranged from those who had never used AI to those with months of daily ChatGPT experience. Pacing that satisfies advanced users bores beginners; pacing that accommodates beginners frustrates experienced users.
Solutions Implemented:
- Tiered homework assignments (basic, intermediate, advanced options)
- Advanced participants encouraged to share discoveries with the group
- NotebookLM provided for self-paced exploration beyond core sessions
- Explicit permission to skip sessions covering familiar topics
- Focus on breadth (many AI applications) creates value even for experienced users who may know text interaction well but haven’t explored data analysis or automation
Challenge 3: Balancing Technical Depth with Accessibility
The Problem: Sufficient depth to build genuine capability risks overwhelming non-technical participants. Excessive simplification leaves participants unable to apply learning independently.
Solutions Implemented:
- “You don’t need to understand how it works to use it effectively” as a guiding principle
- Analogies and mental models prioritised over technical explanations
- Focus on conceptual understanding sufficient for informed decision-making
- Practical application emphasised over theoretical knowledge
- Advanced technical details available through NotebookLM for those interested
Challenge 4: Demonstrating Tangible Value
The Problem: Professional development competes with operational work. Unless programmes demonstrate clear value, participation drops as urgent work takes precedence.
Solutions Implemented:
- Tangible outputs defined for each module (prompt libraries, automated workflows, custom assistants)
- Homework tied to actual work tasks rather than hypothetical scenarios
- Quick wins prioritised in early modules (immediately useful prompt engineering skills)
- Success stories shared regularly (e.g., “I automated this report that used to take two hours”)
- ROI tracking: participants estimated time saved through AI skills acquired
Measuring Success: Programme Outcomes
The programme defines success through four key deliverables participants create throughout the 24 weeks:
1. Prompt Library Development
Each participant builds a minimum of 20 reusable prompts for their specific role. These aren’t generic templates but personalised tools that reflect individual work contexts, communication styles, and recurring tasks.
2. Workflow Automation
Participants design and implement at least one five-step automated workflow that addresses a genuine operational need. Examples from the current cohort include content review pipelines, data extraction processes, and meeting summary generation systems.
3. Custom AI Assistant
Each participant creates a specialised AI assistant configured for their team’s needs, complete with custom instructions, relevant knowledge bases, and specific tone/style guidelines.
4. Data Analysis Pipeline
Participants demonstrate capability to process real business data with AI—uploading datasets, generating statistical insights, creating visualisations, and building data-driven narratives.
These tangible outputs provide both individual capability assessment and organisational benefit. The prompt libraries become shared resources; the workflows improve operational efficiency; the custom assistants serve entire teams.
Beyond these formal deliverables, success manifests in changed behaviour: participants who begin using AI daily rather than sporadically, who help colleagues troubleshoot AI challenges, who propose AI applications for organisational problems, who advocate for AI integration in their departments.
Lessons Learned: What Would I Do Differently
Reflecting on programme delivery reveals several areas for refinement:
What Worked Exceptionally Well
The 30-minute format: This proved the single most important design decision. Participation remained high throughout because the time commitment remained manageable. Professionals protect one hour weekly far more successfully than they protect full-day workshops.
Immediate practice time: Having participants use AI tools during sessions rather than listening to descriptions accelerated capability development dramatically. Those who attended sessions and practised consistently outperformed those who watched recordings without simultaneous practice.
Real-world homework assignments: Tying homework to actual work needs maintained motivation and demonstrated continuous value. Participants completed homework at higher rates when assignments addressed genuine operational needs.
Community learning dynamics: The Teams channel became a vibrant space for peer teaching, troubleshooting, and celebration. Participants answered each other’s questions, shared discoveries, and built collective capability beyond what any individual session could achieve.
What Required Adjustment
Underestimated homework time: Initial homework assignments proved more time-consuming than anticipated. Adjusting to truly achievable 30-60 minute tasks required several iterations.
Advanced module pacing: Modules 4 and 5 (advanced concepts and automation) required more conceptual groundwork before practical application. Future iterations will extend these modules by 2-3 weeks.
Platform diversity: Teaching across ChatGPT, Claude, Copilot, and Gemini created confusion as interfaces and capabilities differ. Future programmes will focus on one primary platform with others introduced briefly for comparison.
Assessment approach: The programme initially lacked formal assessment beyond final deliverables. Adding optional mid-programme self-assessments would help participants gauge progress and identify areas requiring additional attention.
Recommendations for Organisations Implementing Similar Programmes
- Prioritise consistency over intensity: Frequent short sessions beat infrequent long sessions for sustained engagement and capability development.
- Make practice mandatory: Passive observation doesn’t build AI fluency. Every session must include hands-on interaction with actual AI tools.
- Tie learning to operational needs: Generic AI awareness training fails. Programmes succeed when participants solve real work problems using newly acquired skills.
- Build community alongside capability: Individual learning creates individual capability; community learning creates organisational capability. Invest in spaces for peer teaching and knowledge sharing.
- Manage expectations explicitly: AI’s limitations matter as much as its capabilities. Programmes that oversell create disillusionment; programmes that transparently address both strengths and weaknesses build informed, effective users.
- Provide scaffolded support beyond sessions: Learning doesn’t happen only during scheduled sessions. Tools like NotebookLM extend learning environments, creating continuous engagement between formal instruction.
- Define tangible outputs early: Participants need clear targets. Defining deliverables (prompt libraries, workflows, assistants) from the beginning focuses effort and demonstrates progression.
Building Organisational AI Readiness
Effective AI training programmes don’t create AI experts; they build AI-confident professionals who integrate AI effectively into their existing work. The difference is crucial. Organisations need employees who recognise appropriate AI applications, use tools effectively within their domains, understand limitations sufficient to verify outputs, and identify opportunities for AI-enhanced processes.
The 24-week programme detailed here demonstrates that building such capability doesn’t require expensive external consultants, intensive bootcamps that disrupt operations, or technical backgrounds. It requires thoughtful pedagogical design, sustained engagement, practical application, and organisational commitment to long-term capability development rather than superficial awareness.
At Cambridge, this programme serves as infrastructure for organisational transformation. Participants return to their teams as AI advocates and resources, multiplying impact beyond direct programme participation. The workflows they build, the assistants they create, and the prompt libraries they share become organisational assets that compound over time.
As AI capabilities continue expanding, organisations face a choice: adopt AI reactively, struggling to keep pace with change, or build systematic capability that enables proactive integration. This programme represents the latter approach—investing in people rather than just tools, building understanding alongside skills, and creating sustainable capability rather than temporary enthusiasm.
The journey from “AI curious” to “AI confident” takes time, structure, and support. This programme provides all three, demonstrating that with appropriate design, even the busiest professionals can develop genuine AI capability whilst maintaining their operational responsibilities.
About This Work
This training programme forms part of my broader work in applied AI professional development at Cambridge University Press and Assessment’s Education Futures division. The pedagogical approaches detailed here draw on design-based research methodologies, iterating programme design based on participant feedback and outcome analysis.
The 10/10/10 structure emerged from earlier workshop facilitation experience, recognising that working professionals require learning formats that respect time constraints whilst maintaining effectiveness. The focus on tangible outputs reflects product management principles—defining clear success criteria and working systematically towards demonstrable results.
Future programme iterations will extend into specialised tracks (AI for content development, AI for data analysis, AI for project management) that build on this foundational 24-week structure, creating pathways from general AI literacy to domain-specific expertise.
AI Training Programme Examples
- Training Kuwait Teachers – National-scale implementation
- Cambridge AI Leadership Training – School leadership focus
- Teaching Assessment Professionals – Specialized AI training
- AI Curious to AI Confident – Self-paced learning journey