Where to start with AI tools at work
The most common mistake with AI at work is starting with the wrong question: "how do I use AI?" instead of "what takes me the most time and produces the most commodity output?" Start by identifying the tasks in your work that are high-effort but relatively formulaic: drafting emails that follow a standard structure, summarising long documents, creating first drafts of reports, preparing meeting agendas, writing job adverts or policy documents, researching standard background information. These are the tasks where AI provides the most immediate return and where errors are easiest to catch.
Do not start with high-stakes, high-judgment tasks (advising clients, making business decisions, writing formal legal or financial documents without review) until you have developed good calibration for when AI output is reliable and when it is not. AI tools make confident mistakes. Developing the judgment to spot them requires practice with lower-stakes material first.
Which AI tools to use in 2026
The main AI tools relevant to most knowledge workers: Claude (Anthropic) — strong for long-document analysis, nuanced writing, and professional communications. Handles long context windows well. ChatGPT (OpenAI) — broad capability, good code generation, web search integration in paid tier, large plugin ecosystem. Microsoft Copilot — integrated into Office 365, Teams, Outlook, and Word; particularly valuable if your organisation already uses the Microsoft stack because it can access your own documents and emails. Google Gemini — integrated into Google Workspace; useful for organisations using Google Docs, Gmail, and Sheets. For most professional use cases, the practical differences between leading models are smaller than the marketing suggests. Start with the tool your organisation already pays for (Copilot in most enterprise contexts), and try others for specific tasks where you want comparison.
How to write prompts that get useful results
Three things make the biggest difference to AI output quality: Context: tell the AI who it is writing for, what the purpose is, what constraints apply, and what the output should look like. A prompt that says "write a summary of this meeting for a non-technical audience that should be no longer than one page" produces far better output than "summarise this." Examples: if you want output in a specific format or style, show an example. Paste in a previous email you wrote that struck the right tone and say "write in this style." Iteration: treat the first output as a draft, not a final product. Follow up with specific refinement: "make the second paragraph shorter," "the tone is too formal — make it more direct," "add a specific example to the opening." This iteration loop is where AI becomes genuinely useful for professional work.