AI Foundations Companion

Start here: Your foundation in AI

1.1 Welcome – moving from overwhelmed to empowered

Many people feel overwhelmed by AI. In my workshops, I emphasise that curiosity and experimentation matter more than expertise. AI is a practical tool – not a magic wand. We are all learning together, and the model you use today is the slowest and least capable it will ever be. That's exciting because improvement is inevitable.

1.2 Core concepts – what you actually need to know

AI is a simulation of intelligence. It processes vast amounts of data to generate plausible responses. It doesn't think or understand – it predicts. Knowing this helps you treat AI as a powerful but fallible tool, not a sentient colleague.

• Generative AI & LLMs. Modern systems like ChatGPT and Gemini are built on Large Language Models (LLMs). They generate text by predicting the next most likely word. This allows them to draft, summarise and converse in natural language.
• Reasoning vs non-reasoning models. Some models are optimised for speed, producing quick responses without deep reasoning. Others can be instructed to pause, check their work and follow complex logic. Use fast models (e.g. Gemini Flash, GPT-4o) for simple tasks and reasoning models (e.g. Gemini Pro, GPT-4 o3 or o4) when accuracy matters.

• The "jagged edge" of AI. LLMs excel at some tasks and fail at others. A model may summarise a 30,000-word report brilliantly yet miscount bullet points in its own output. Always test on your own tasks and apply human judgement.
• Retrieval-Augmented Generation (RAG). RAG connects a model to trusted documents or databases, allowing it to "chat with your data." This yields more relevant answers because the model can reference your own content.
• AI agents. An agent is an AI system that can take a goal and independently execute a multi-step plan. For example, I used a research agent to investigate wind turbine foundation settlement values and produce a cited report in minutes. Agents are promising but still require oversight.

1.3 Jargon buster

• AI – Any system that performs tasks requiring human-like intelligence.
• LLM (Large Language Model) – A model trained on large text datasets that predicts the next word.
• AGI (Artificial General Intelligence) – A hypothetical AI that can perform any intellectual task a human can. We are not there yet.
• RAG (Retrieval-Augmented Generation) – Combining AI with your own documents/data for customised answers.
• Agent – An AI system that can plan and execute a series of steps toward a goal.
• Co-pilot – An AI embedded in software (e.g. Microsoft 365 Copilot) that assists with tasks.

A practical AI toolkit

2.1 The essential platforms – your "big three" and Copilot

The key is to choose the right tool for your task, as each platform has different strengths. You don't need to master them all; focus on one or two that fit your workflow.

• Google Gemini. Excellent at deep research and handling very large documents. Its ability to produce structured outputs like tables and its integration with Google's ecosystem make it a powerful choice for analysis.
• OpenAI ChatGPT. A great all-rounder with strong conversational abilities and an advanced voice/vision mode that can interpret what it sees through your phone's camera. Its main advantage is a vast ecosystem of custom assistants (GPTs) for specialised tasks.
• Anthropic Claude. Known for its nuanced and natural writing style. It excels at summarising long, complex documents and creative writing tasks.
• Microsoft Co-Pilot. Its primary value comes from being integrated directly into Word, Excel, Outlook, and Teams. This makes it easy to adopt for tasks inside the Microsoft 365 environment.

2.2 Specialised tools for specific tasks

Beyond the main platforms, a number of specialised tools are useful for specific jobs.

• Image generation (Midjourney). Produces high-quality, professional images from simple text prompts. Useful for creating unique visuals for presentations or marketing materials.
• Voice synthesis (11 Labs). Generates realistic, human-sounding voices from text. This is useful for creating audio for training videos or internal communications.
• Meeting transcription & analysis (Otter.ai, Fathom). These tools automatically record and transcribe calls. Practical Workflow: Feed the raw transcript into an AI like Gemini or Claude to create structured meeting minutes, identify action items, and summarise key decisions.
• Document analysis (Google NotebookLM). This tool lets you "chat with your documents." Practical Workflow: Upload a long technical report and ask it to generate an audio overview for a new hire or create a mind map of the key concepts.

2.3 Calculator vs. Creative Partner

LLMs are probabilistic - they guess the next most likely word. They are not calculators. If you ask them to perform arithmetic, they can make mistakes. The winning move is to treat the model as a creative partner. Ask it to build a deterministic tool - like a formula or a script - that you can then use for accurate calculations. This combines the AI's creativity with the precision of traditional software.

Prompting: How to Instruct and Talk with AI

3.1 Why prompting matters

The quality of the output you get from an AI is determined by the quality of your input. A vague prompt will yield a vague answer. A clear, structured prompt will yield a better, more useful result. This is important because it puts you in control. Think of prompting like briefing a colleague: be explicit about the goal, audience, length, tone, and provide examples if you can.

3.2 Three styles of prompting

There are three main ways to interact with an AI.

• Simple prompting. A direct question or a single instruction. Use it for quick facts or straightforward tasks like "Translate 'hello' to French."
• Conversational prompting. A back-and-forth dialogue where you provide more context, ask follow-up questions, and refine the output together. This is excellent for brainstorming or working through a complex problem.
• Structured prompting. A detailed set of instructions with clear sections for the goal, output format, and context. Use this for recurring tasks where you need consistent results every time.

3.3 The GOAL framework

For structured prompts, I use a simple four-step framework to make sure I cover all the essential information.

• G – Goal: Define exactly what you want to achieve.
• O – Output: Specify the desired format, length, tone, and audience.
• A – Additional context: Provide relevant background information, examples, or constraints.
• L – Look over: Review the AI's output, spot any errors or gaps, and iterate on your prompt.

AI safety, privacy and ethics

4.1 You are 100 percent responsible

AI systems are designed to produce plausible-sounding answers, not necessarily truthful ones. They can and will make mistakes (sometimes called "hallucinations"). They also reflect the biases present in their training data and can be overly agreeable or flattering ("sycophancy"). This is important because you cannot outsource accountability. You are always responsible for checking, correcting, and owning the final output. If you send a report or email that was drafted with AI, you are the one sending it. Your professional reputation is on the line.

4.2 Data privacy – free vs enterprise

This is a critical distinction. Never put sensitive client or company data into free, public AI tools. Those services often use your data to train their models, meaning your information is not private. Always use your company's paid or enterprise account (e.g., ChatGPT Team, Gemini for Workspace). These versions typically include contractual data protections that prevent your information from being used for general model training. When you want to experiment with a new, unapproved tool, create a "synthetic client" with realistic but entirely fake data to avoid exposing real information.

4.3 Policy and disclosure templates

When you share AI-assisted work externally, it's good practice to be transparent.

Internally, it's useful to have a simple AI usage policy for your team. It doesn't need to be complicated.

From theory to practice: your implementation roadmap

5.1 Find your use case – the RATES framework

Knowing the concepts is one thing; applying them is another. To identify where AI can help you most in your own work, I recommend a simple framework called RATES. It helps you score your daily tasks to find the best candidates for automation.

• Repetitive – Tasks you do over and over, like drafting weekly reports.
• Annoying – Tasks you dislike, such as formatting documents or transcribing notes.
• Time-consuming – Tasks that take a long time, like cleaning data or conducting initial research.
• Error-prone – Tasks where small mistakes can be costly, like copying numbers between spreadsheets.
• Scalable – Tasks where improvements would benefit many people, like creating better onboarding materials.

The best place to start is often with a task that is both repetitive and annoying. Removing these from your workflow provides an immediate and motivating win.

5.2 Build single-purpose assistants

Once you find a recurring task, you can create a custom assistant to handle it. In ChatGPT, these are called Custom GPTs, and in Gemini, they are Gems. The key is to build single-purpose assistants rather than one "master" AI that tries to do everything. A dedicated assistant with a clear role and a structured prompt will always produce more reliable and consistent results.

5.3 The 4-week action plan

This roadmap offers a structured approach to integrating these concepts into your work.

• Week 1 – Quick wins:
  • Log in to your company-approved AI tools.
  • Use an AI to summarise a report or a long email thread.
  • Practice using the GOAL framework for one simple, repetitive task.
• Weeks 2–3 – Exploration:
  • Block 30 minutes, twice a week, in your calendar for "AI Experimentation."
  • Try a new feature, like the transcription workflow or using NotebookLM with a document.
  • Attempt to automate one part of the high-scoring task you identified with RATES.
• Week 4 – Integration:
  • Develop a detailed, personal structured prompt for a common task you perform.
  • Try building your first simple Gem or Custom GPT for that task.

5.4 Your very next action

Momentum builds competence. Within the next 24 hours, use an AI tool to complete a small part of your work. It could be drafting the introduction for a report, brainstorming three subject lines for an email, or checking your grammar on a document. Evaluate the result and edit it yourself.

Working with your data (RAG)

RAG, or Retrieval-Augmented Generation, is a technique that combines a powerful AI model with a secure, private knowledge base. Instead of relying only on its public training data, the model can fetch information from your documents to answer questions.

This is important because it makes the AI's answers highly relevant and accurate to your business. It's how you get an AI to answer questions based on your internal world, not just the public internet.

A great way to start is with a small, focused pilot project. Build a simple "document chat" for one team.

AI Agents and what is possible today

AI agents are systems that can take a goal, create a plan, and then execute that plan with multiple steps. They can conduct research, write drafts, and in some cases, interact with software.

However, they are not fully autonomous. Agents can still make mistakes, get stuck on a step, or follow incorrect instructions. This is important because you still need to set clear goals, monitor their progress, and review the final output. Human intervention remains essential.

When an agent makes sense vs. a simple assistant

Knowing when to use an agent versus a simpler custom assistant (like a Gem or GPT) is key.

• Use an agent for:
  • Multi-step tasks with a clearly defined, objective goal (e.g., "Compile a market landscape report on our top three competitors").
  • Broad research tasks where you need to gather information from multiple sources.
• A simple assistant plus a checklist is better for:
  • Highly specific tasks that require your professional judgement or nuanced context at each step.
  • Tasks that involve sensitive data or require access to local systems where you want full control.

Use Cases and Recipes

Once you are comfortable with the tools, you can apply them to specific functions in your business. Below are some ready-made workflows. I encourage you to adapt them to your own documents and needs.

Sales: proposal from call transcript

This workflow can dramatically speed up the time it takes to follow up with a potential client.

1. Record your call with the client (always get consent first).
2. Transcribe the call using a tool like Otter.ai or Fathom.
3. Feed the transcript into an AI with a structured prompt.
4. Review, edit, and send the AI-drafted proposal.

Marketing: repurpose assets

This is a powerful way to get more value from content you've already created.

1. Provide a source document, like a blog post or a webinar transcript.
2. Prompt the AI to create derivative assets for different channels.
3. Fact-check and adjust the messages to fit each platform.

Operations: SOPs from meetings

Use AI to turn conversations about processes into clear, documented procedures.

1. Record a meeting where your team discusses a specific process.
2. Transcribe the meeting and feed the transcript to an AI.
3. Share and refine the AI-generated Standard Operating Procedure (SOP) with your team.

HR: job descriptions & interview rubrics

Speed up the hiring process by getting AI to handle the first draft of key documents.

1. Supply the AI with the key role requirements and notes on your team culture.
2. Ask the AI to draft the job description and a structured interview rubric.
3. Check for bias and accuracy before posting or using in an interview.

Adoption, leadership and jobs

Many people worry that AI will replace jobs. In my talks, I advocate for a different approach: using AI to do more with the same people, not the same with fewer. Choosing to cut staff destroys institutional knowledge and valuable relationships. The better strategy is to use AI to amplify your team's capability, freeing up time for higher-value work like strategy, client relationships, and innovation.

There is often a gap between how employees use AI and how much leaders are aware of it. Staff might quietly automate parts of their jobs without telling managers for fear of being misunderstood or having their workload increased. To close this gap and normalise AI use, leaders should:

• Use AI themselves. Hands-on experience builds understanding and credibility. When leaders use the tools, it signals that experimentation is encouraged.
• Encourage sharing. Allocate time in team meetings for people to demonstrate their "best AI use" of the week. This makes learning a team activity and helps good ideas spread.
• Provide training. Offer practical workshops and resources (like this one) to upskill people and build a common language around AI.

How to Measure Value: Four Pillars and a simple ROI

10.1 Four Pillars of AI impact

To understand the value AI brings, it's useful to think beyond just time saved. I encourage people to measure the impact of AI across four key pillars:

• Productivity. This is the most obvious one - time saved and efficiency gains on specific tasks.
• Capabilities. This is about doing work that was previously impractical or impossible. For example, I'm not a graphic designer, but I can now generate high-quality, on-brand images for my content in minutes.
• Decision making. AI can improve the quality of your decisions by processing large amounts of information, identifying patterns, and helping you analyse complex scenarios.
• Learning. AI can dramatically accelerate learning and skill development. It allows you to get up to speed on new topics quickly and can be used to create effective onboarding and training materials.

This is important because focusing only on time saved misses much of the strategic value AI can offer. Expanding your team's capabilities or improving decision quality can have a far greater impact than simply doing the same tasks a bit faster.

10.2 A simple ROI tracker

To make this practical, you can use a simple spreadsheet to track the value you're getting from AI. This doesn't need to be complicated, but it helps to make the benefits visible.

Use this to track a few key tasks over a couple of weeks. The results can be a powerful way to show the value of AI to your team and to leadership.

Frequently Asked Questions

Here are short, practical answers to some of the most common questions that come up in my workshops.

Further Resources

This final section provides links to downloadable templates and other useful materials mentioned throughout this guide.

• GOAL Prompt Card: A one-page PDF summarising the GOAL framework with a clear example prompt.
• Meeting Minutes Kit: Includes consent language for recording calls, a pre-meeting checklist, and a structured prompt for generating minutes from a transcript.
• RATES Worksheet: A simple spreadsheet to help you log and score your daily tasks to find the best automation candidates.
• ROI Tracker: A spreadsheet template for measuring the impact of AI across the Four Pillars.
• Calculator vs. Creative Partner Explainer: A short document with examples and prompts for asking AI to build formulas and scripts for you.
• Links to my public Custom GPTs and relevant articles will be maintained on a dedicated page on my website.