Using AI right now. How I’m doing it, and how you can too.
I spend my days helping Kiwi business leaders understand how to practically apply AI in their work. So, here’s my take on which AI tools I use, how I use them, and what I’ve found works best.
The big four AI assistants I use every day
If you’re serious about using AI for work, I recommend starting with one (or more) of these:
ChatGPT (OpenAI) – excellent for structured outputs, coding, and creative content. The new GPT-4o model is incredibly fast and capable.
Claude (Anthropic) – great for long documents, deep thinking, and privacy. Claude won’t train on your data.
Microsoft Copilot – seamlessly integrated with Microsoft 365 apps like Word, Excel, and Outlook. It’s powerful for drafting documents, generating summaries, and automating admin across the tools many businesses already use.
Perplexity – my go-to for real-time research. It’s like search on steroids, built for professionals. It doesn’t just answer questions; it gives sources and structured summaries.
Each has its strengths, and I rotate between them depending on the task. For example, I might use Claude to summarise a report, Copilot to draft slides in PowerPoint or run data analysis in Excel, ChatGPT to rewrite it for different audiences, and Perplexity to validate the data and find citations.
Choosing the right model for the job
Each platform offers multiple models, fast, powerful, and sometimes ultra-powerful. For real work, you’ll want to switch to the most capable version available:
ChatGPT: use GPT-4o (or GPT-o3 for advanced reasoning and deep outputs)
Claude: switch to Opus
Microsoft Copilot: built on OpenAI’s models, typically running GPT-4, and optimised for tasks inside Word, Excel, PowerPoint, Outlook and Teams
Perplexity: uses strong retrieval-augmented models by default, so you’re usually good to go
Free versions are fine for testing, but if you’re using AI seriously, expect to pay around US$20–40/month to unlock full capabilities.
Privacy-wise, Claude is the most conservative, it never trains on your data. For ChatGPT and Copilot, you can turn off training in the settings (in Copilot’s case, business users are usually covered by enterprise-grade data policies). I always disable training where I can.
Here are the four characteristics of a business processes ripe for AI automation
Over time, I’ve learned to identify business processes that AI can dramatically enhance. Here’s the framework I share with clients:
Search – When finding accurate information is slow or hard - E.g. researching competitors, regulations, or market trends.
Serve – When your team needs to give instant, consistent answers - E.g. providing product info to customers or contractors needing guidance during site inspections.
Create – When you’re turning complex data into structured outputs - E.g. writing RFPs, consent applications, or business reports.
Enhance – When business-critical decisions rely on fragmented data - E.g. resource allocation, predictive maintenance, logistics planning.
The most underused superpower: Deep Research
Most people don’t realise AI can generate high-quality reports that feel like something a consultant would charge thousands for.
I use Deep Research to:
Write strategic market overviews
Build competitive analysis decks
Generate CEO briefings on how AI is impacting their industry
Perplexity is brilliant here because it cites its sources and lets you dive deeper. Claude and ChatGPT also do solid work, especially when you turn on web browsing. Gemini’s reports can even turn into quizzes or podcasts with one click.
Voice mode is more powerful than it looks
I often use AI while walking or commuting. Voice mode in the ChatGPT and Gemini apps lets me have natural conversations and think through challenges aloud. But the real magic is multimodal, point your phone at a broken part, a whiteboard, a sign in another language, and the AI will help in real time.
If you’re just using voice mode like Siri, you’re missing the best part.
Making things with AI: code, docs, images, and video
ChatGPT is my favourite for image generation and structured outputs like documents, dashboards, and workflows. Gemini is solid too, and Veo (its video model) creates surprisingly slick results with little effort. Claude can’t make images, but it’s excellent at writing, summarising, and analysing.
Want AI to write a policy doc, turn it into a presentation, then create a social post and podcast episode about it? You can do that right now.
Prompts aren’t magic anymore, context is
The key shift I see with today’s models: it’s less about the prompt and more about the context.
Upload a doc, explain your situation, and be specific. Instead of “write me a job ad,” say “write a job ad for a construction site supervisor in Auckland, here’s our tone of voice, key responsibilities, and company story.” Then refine from there.
Ask for variations. Explore options. Use branches. Push it to do more, it doesn’t get tired.
My recommendation: spend your next hour like this
Here’s how I onboard new clients or team members into AI:
Pick a system and pay for it, the free versions are demos.
Switch to the powerful model, then give it a real task from your job. Add full context. Ask for changes until you love the output.
Try Deep Research on something you care about, a strategic issue, a product launch, or a market you’re entering.
Experiment with voice mode while doing something else. Let the AI think with you in real time.
Most people use AI like Google: quick, surface-level, disposable. But when you give it depth, real work, real context, it’ll surprise you.
