Skip to content
AI New Zealand
Our Playbook Find your Benchmark Executive AI Briefing AI Lunch & Learn AI Operating System Client Workspace Book Justin to speak AI Courses: Certify Your Team AI for schools: Schools & Students AI at Work Foundations course AI Leaders course Claude at Work course Microsoft Copilot course ChatGPT at Work course Gemini at Work course Putting Descript to work course Putting Midjourney to work course AI Roadshow Premium membership Toolkits Webinars Articles & Insights AI Academy Contact us
AI at work

The blank page is now optional.

What changes when you start with what you know, not with Word.

A blank white page with a gold cursor, beside the headline: The blank page is now optional

When did you last start a document with a blank page? For most of us, it was probably this week. A new Word document, a blinking cursor, and the slow work of building something from nothing.

That's how office work has run for decades, and for most businesses it still does. But a growing number of people are changing what they expect. They're not starting with a blank page any more. They're starting with what they know and what they need, and letting the document take shape from there.

The recent Microsoft 365 outage made me think about this more. It started early on Friday 13 November, New Zealand time, and ran for more than a day. Nearly 130,000 users reported problems with Outlook on Downdetector, and Microsoft confirmed the issue could affect users in any region. It wasn't just email either. Teams, SharePoint and Copilot were among the services Microsoft listed as affected. Microsoft later apologised and put it down to a technical problem during a network capacity upgrade, not a cyberattack.

Every provider has bad days, so this isn't about Microsoft being unreliable. But it was a reminder of how much of our working day runs through one place, just as the way we work in that place is changing. Add a few weeks spent with organisations that are quietly changing how they work, and here's what I'm thinking.

At a glance

  • Work is shifting from building documents piece by piece to describing what you need and letting AI produce it.
  • Microsoft sees this too, and is building towards it. This isn't a story about Microsoft losing.
  • The costs of using AI properly add up in different ways depending on the tool, so compare what you'd actually use, not just the seat price.
  • Using more AI tools brings real security and privacy questions, and where your data lives is less clear-cut than most people think.
  • The best place to start isn't choosing a tool. It's understanding how your work gets done.

I spend a lot of my time in Microsoft. At AI New Zealand we've trained thousands of people in Copilot, Microsoft's AI assistant, and we work with hundreds of organisations on getting real value from the AI tools inside Microsoft 365. We also do a lot of work in Claude, an AI assistant made by Anthropic, and we see where ChatGPT and Google's Gemini turn up in our clients' businesses. So this isn't about one vendor beating another. It's about a change we're seeing across all of them.

What's actually changing

Here's the simplest way I can put it.

The old way: you need a one-page summary of last month's sales for the bank. You open Excel, pull the figures together, build a chart, open Word, write the summary, paste the chart in, fix the formatting and save it as a PDF.

The new way: you tell an AI tool what you need. Something like “turn last month's sales figures into a one-page summary for the bank, with a chart and a short explanation of why March was down”. It produces a draft. You check the numbers, correct what's wrong, add what only you know, and send it.

You haven't stopped doing the work. You've stopped doing the assembly. Your job moves from building the document to knowing what it needs to say and checking that it says it properly.

Notice what's missing from the new way. There's no blank page.

That's not a lesser skill. If anything it asks more of people. You need to know your subject well enough to spot what's wrong, and know the business well enough to recognise what good looks like. The typing has become the easy part.

How we work now

At AI New Zealand, we don't start a document from scratch. Not a proposal, not a workshop pack, not a report.

Everything starts as a conversation with Claude and becomes a shared working draft. People in the team edit it directly, comment on it and pass it around. The thinking happens there: what the client needs, what we know, what the right answer is. Nobody's worrying about the format yet.

Only at the very end do we decide what it becomes. It might be a Word document, a PDF, a slide deck, an email or a web page. At that point our rules, templates and branding get applied, and the finished piece is produced in whatever shape the person receiving it needs.

So the format is the last decision, not the first. We start with the information and what we're trying to achieve, and the document comes out the other end.

What this looks like in a smaller business

You don't need to be a tech company to work this way. Picture a 15-person engineering or trades business. Every week someone writes quotes, a site report or two, and a summary for the owner.

Today, each of those starts with a blank template and an hour of typing. Working the new way, the person feeds in the job notes, photos and figures, describes what's needed, and gets back a first draft in the company's usual format. They spend their time checking the detail and adding judgement, not formatting tables.

A growing number of the organisations we work with are heading this way. Plenty still start in Word, and that's fine. But the ones moving fastest are starting with the information instead.

How we got here

For thirty-odd years, office work meant building things piece by piece. You wrote the document. You built the formulas in the spreadsheet. You laid out the slides one text box at a time. If you needed to show data, someone built a report in a tool like Power BI, and everyone else learned where to click.

Microsoft owned that world, and it owned it well. Word, Excel and PowerPoint became the default, and their files became the standard way businesses swapped information. Your email, files, logins and security all lived in the same place. Choosing Microsoft meant a lot of decisions were made for you, including many of the rules about who could see what.

Microsoft also saw the next change coming earlier than most. It invested heavily in OpenAI, the company behind ChatGPT, and made Microsoft 365 Copilot generally available to large businesses on 1 November 2023, putting AI inside the apps people already used every day.

In September 2025 Microsoft went further and brought Anthropic's Claude models into Microsoft 365 Copilot alongside OpenAI's. It has since extended them into Copilot inside the Office apps themselves. I read that as a smart move and a telling one. To me it says Microsoft cares less about whose AI does the work than about being the place where the work happens.

Why the file matters less

Word and Excel files use open formats, which means any software can create them. More and more, it isn't Office doing it. The bank still wants a spreadsheet and the auditor still wants a document, so those files aren't going anywhere. But the program that made them matters less than it used to.

And the end result may not be a document at all. Where someone used to build a report and share a link, it's now possible to ask a question of your data and get back an interactive web page made for that question. We do this in our own work, and we're starting to see clients try it.

There's a catch worth being honest about. A good reporting tool isn't valuable because of its charts. It's valuable because everyone gets the same trusted numbers, the right people see the right information, and you get the same answer every time you ask. A report created fresh each time can give a different number on a different day if the data underneath is messy. So the trick is to keep your core data tidy and trusted, and let AI do the presenting on top of it.

Microsoft sees it too

It would be easy to say Microsoft is being left behind. That isn't what's happening.

Microsoft is building towards the same future. A feature called Copilot Cowork lets you hand Copilot a task and have it carry the work through, rather than just answering a question. Agent 365 is a way for IT teams to keep track of, and control, the AI helpers an organisation runs. And Microsoft's top-tier business licence, Microsoft 365 E7, adds Copilot, Agent 365 and the Microsoft Entra Suite of identity tools to everything in Microsoft 365 E5.

An analysis in SiliconANGLE earlier this year put the challenge well: to own the future of work, Microsoft may need to let Office become less central, so it can own whatever replaces it. That's a hard thing for any company to do with its most successful product, and Microsoft is clearly working through it.

Microsoft's strengths in logins, security, compliance and data are real too. As AI spreads through a business, those things matter more, not less.

What about ChatGPT?

In our client work, ChatGPT is mostly used as a general-purpose chat assistant. OpenAI's technology shows up a lot behind the scenes, built into other products. What we haven't seen much of is ChatGPT being used day to day to produce a business's documents, alongside or instead of Microsoft. That's where the real contest is right now.

What does it cost?

Cost is usually where business owners start paying attention, so it's worth comparing like with like.

The short version: the price per person is similar. The difference shows up once your team starts using AI heavily.

At US list prices, paid yearly, Microsoft 365 Copilot Business is US$21 per person per month, or US$18 for the first year on a promotion for existing Business plan customers that runs until 31 December 2026. It's an add-on, so you still need a Microsoft 365 plan. Claude's Team plan is US$20 per person per month, and most businesses using it still keep Microsoft 365 for email and files.

The bigger difference is in getting AI to carry out tasks or build helpers. With Microsoft that's often charged extra, by usage, and heavy users buy more usage credits. Copilot Cowork tasks use Copilot Credits, and building custom AI helpers in Copilot Studio beyond the basics also runs on credits, sold in packs of 25,000 for US$200 a month.

With Claude, tasks and helpers are included up to a usage limit, and beyond that you pay for extra usage or move up a plan. It isn't unlimited either. Every Claude plan has usage limits, and heavier use costs more. Its larger Enterprise plan charges for all usage on top of the seat price.

Neither is automatically cheaper. It depends entirely on how your people work. One note, because the names confuse people: Microsoft's Copilot Cowork and Anthropic's Claude Cowork are different products from different companies.

Our clients are feeling this at renewal time. One told us recently that their last Microsoft renewal came with a significant increase, and they're expecting more. That's when people stop accepting the bill as a given and start asking what they're actually paying for.

Scammers are using AI too

Any conversation about using more AI tools has to cover security, and early.

AI is making scam emails faster to produce and harder to spot. Microsoft's own 2026 Digital Defense Report found phishing, where a fake email tricks someone into handing over access, accounted for 23% of the intrusions it observed in 2026, up from 7% in 2025. Email security firm Cofense saw, on average, one malicious email every 19 seconds in 2025, more than double its 2024 pace of one every 42 seconds. And KnowBe4 reported in April 2026 that 86% of phishing attacks over the previous six months were AI-driven.

In practice, the old warning signs, the bad spelling and the “Dear Customer” greeting, are fading. Convincing scam emails can now be aimed at far more people for far less effort.

That's why it matters how many tools you use. Every new tool you connect is another login, another set of permissions and another place your information can be reached. If scammers are getting better at finding the weakest link, more links raise the stakes.

Microsoft is strong on security, and for many businesses it rightly stays the foundation. Plenty also use a specialist security company, because security is all that company does. The same thinking is starting to apply to AI: a specialist for AI, a specialist for security, and Microsoft for email, logins and files.

That can work well, but only if someone is responsible for how it all fits together. For years, using only Microsoft did a lot of that quietly in the background. Step outside it and that job becomes yours.

Where is your data, really?

This is the question New Zealand businesses ask us most, and the answer is less clear-cut than most people assume.

Start with Microsoft. Microsoft now has datacentres in New Zealand, and Microsoft 365 data can be stored here, including email, SharePoint, OneDrive, Teams and Copilot. That's a real step forward. But the guarantee that it stays here depends on how your account is set up and, for some services, on an extra licence such as Microsoft's Advanced Data Residency add-on, so it's worth checking rather than assuming.

It gets more complicated once AI is involved. Copilot can now use Anthropic's Claude models, and for most New Zealand business accounts they're switched on by default, under Microsoft's terms, with Anthropic working as one of Microsoft's subprocessors. But Microsoft says Claude currently sits outside its in-country processing commitments. And some of Anthropic's newest models run under Anthropic's own terms, with Anthropic keeping data for up to 30 days, once an administrator switches them on. So the same Copilot session can handle your data differently depending on which AI model is doing the work.

Claude has its own picture. Anthropic offers local data options in several countries, including Australia but not New Zealand, through cloud platforms such as AWS Bedrock and Google Vertex, and what's available varies by platform. On its business plans, Anthropic doesn't use your content to train its AI by default.

Then there's the law, which surprises people. New Zealand's Privacy Act doesn't require personal information to stay in the country. According to the Privacy Commissioner, the rule on sending information overseas doesn't stop you using a cloud provider that stores or processes information on your behalf. The provider is treated as holding it for you.

That doesn't mean anything goes. You're still responsible for keeping the information secure. And if a provider uses your information for its own purposes, it's no longer just holding it for you. Sharing it with them can then count as a disclosure, which the Privacy Act's rules have to allow, and if the provider is overseas the rule on sending information offshore may apply too. That's why settings like AI training, how long data is kept and which AI models are switched on matter as much as where the datacentre is.

So the honest answer to “where is our data?” is usually “it depends on the tool, the plan and the settings”. I'm not a lawyer, and businesses holding sensitive information should get proper advice. But you can start with three questions for whoever looks after your IT:

  • Which AI tools are our people using, including ones we haven't officially approved?
  • For each one, where is our information stored, and is it used to train the AI?
  • Who in our business is responsible for checking those settings, and when did they last look?

Where this is heading

I don't think Microsoft is going anywhere. It's more likely to become the foundation for email, logins, security and files that many other tools sit on, while the place where people actually do their work becomes more contested.

This will take time. Microsoft is deeply embedded in New Zealand businesses, and that changes one decision at a time. But we see the direction in our client work every week.

The real contest is over where people go to say what they need. That might be Copilot, Claude or something we haven't seen yet.

Where to start

The obvious response to all this is to start comparing products. I'd start somewhere else.

What we see again and again is businesses wanting the new kind of output without the foundations underneath. Their information is scattered and out of date. Nobody has set rules for which AI tools are fine to use. And nobody has a clear picture of how the tools, the information and the people fit together.

So start small and practical:

1 Pick one document your team produces regularly

A weekly report, a standard quote, a monthly summary. Something that takes real time and follows a familiar pattern.

2 Write down how it gets made today

Who does it, where the information comes from, how long it takes and where it goes wrong.

3 Try producing it the new way

Describe what's needed, feed in the information, and check what comes back. Note what it got right and what still needed a person.

4 Look at what that tells you

Is your information good enough to rely on? Are the right people checking the output? Is the tool set up safely?

That one exercise will teach you more about your readiness than any product comparison. It shows you where your data, your processes and your people stand.

Then put real effort into the people side. Moving from building documents to describing outcomes changes how people work day to day. Switching on a licence doesn't make that happen. It takes support, practice and time.

Once that's clear, the technology choice gets much easier. It stops being about which vendor wins and becomes about which tools fit the way your business works, and how they fit together safely.

How this article was written

I started by asking when you last began a document with a blank page. I didn't, for this one. I didn't touch a keyboard at all.

I wrote this article by talking to Claude, over about 90 minutes to two hours. I talked through what I'd been seeing with clients. I argued with its suggestions when I thought they were wrong, and it pushed back on mine. I asked it to check facts against the original sources, to find the holes in the argument, and to read the draft the way a sceptical business owner would. We went back and forth, fleshing out the ideas that held up and cutting the ones that didn't.

Honestly, it felt more natural than writing. Talking things through, thinking out loud and debating an idea are some of the most natural things people do. That's where I brought the most: my experience, my judgement and my opinions. The AI brought what it does well: following a natural conversation, looking things up straight away, and turning the thinking into structured writing.

I wasn't writing a document. I was having a conversation, and the document came out the other end.

If you want to try it, pick something you'd normally sit down and write this week, and talk it through with an AI tool instead. See how much of the work turns out to be the thinking.

The format is the last decision, not the first.

Start with what you know and what you need, and let the document take shape from there.

Power up your potential with practical AI skills.

Contact us to discuss how Artificial Intelligence could boost your business.