Browse by topic

Copy, campaigns and content that came from research

Applying AI to marketing work: landing page copy, content planning, ad testing and audience research — grounded in real customer input rather than generated averages.

Writing instructions that get reliable results

The craft of getting useful, repeatable output from a language model: structuring a request, supplying context, constraining the format, and knowing which of those actually moves the result.

How these systems actually work

Transformers, training, fine-tuning, temperature and the mechanics underneath the interface — explained for people who use these tools rather than build them.

Cleaning, analysing and presenting information

Using AI on real data: cleaning messy spreadsheets with rules you can audit, choosing charts that inform rather than mislead, and building views that answer a specific question.

What AI actually costs to run

Tokens, pricing, latency and context limits — the operational numbers that decide whether an AI feature is viable, and how to bring them down.

GPT models and the OpenAI platform

OpenAI's models and tooling — GPT, Custom GPTs, the API and the assistants platform — and how they compare with everything else available.

Using models on real code without shipping what you do not understand

Debugging, reviewing and refactoring with AI assistance — and the discipline of verifying generated code before it reaches a codebase you are responsible for.

Wiring AI into work that runs without you

Putting models into real workflows — content pipelines, triage, data processing — with n8n and similar tools, and knowing which steps should stay human.

Anthropic's model family

Working with Anthropic's Claude models: where they differ from the alternatives, how they respond to instructions, and what they are genuinely better and worse at.

Which model for which job

Honest comparisons across the major models: benchmarks, real-world differences, and the far more useful question of which one to reach for on a given task.

Giving models the information they need

Retrieval-augmented generation, context windows, embeddings and vector search — the techniques for getting a model to answer from your documents rather than its training data.

Knowing when a model is making it up

Why models produce confident, fluent, wrong answers — and the prompting and verification techniques that reduce it. The most under-taught skill in AI use.

Meetings, planning and the admin that eats your week

The unglamorous wins: turning rambling notes into decisions and owners, planning a day that survives contact with reality, and drafting the messages you keep rewriting.

Models that take actions, not just answer

Agentic systems: tool use, function calling, multi-step loops, and the practical question of when handing a model autonomy is worth the loss of predictability.

Proving you can actually use this

Applications, interviews, portfolios and skills — using AI well enough to show it, in a market where everyone claims it and few can demonstrate it.

Llama, Mistral, DeepSeek and running models yourself

Open-weight models you can run, fine-tune and self-host: what they cost, where they match the closed alternatives, and where they genuinely do not.