A luminous globe and glass news panels at opposite sides of a blue background.

Weekly Highlights — 22 November 2026

15–21 November 2026 · New Zealand time

Ten developments in AI, automation and digital technology, with a New Zealand business perspective.

This week, Anthropic went public, OpenAI and Google released new models, and Microsoft and Google showed two ways AI can take on recurring or multi-step work. There were also updates worth watching for WordPress site owners, designers and publishers.


A stock-exchange building beside an illustrated rising chart.

AI industry

1. Anthropic starts trading on Nasdaq

The company behind Claude began trading on Nasdaq on 19 November under the ticker ANT and announced that its IPO had closed the following day. The listing gives customers, competitors and researchers another public record to follow as major AI providers invest in computing capacity and expand their services.

Sources · Anthropic · Reuters


Two computer chips, one blue and one amber.

AI models

2. OpenAI launches GPT-6.2 Astra and Sol

OpenAI released two models on 17 November. It positions Sol for everyday work and Astra for more demanding projects, with both available through the API. If you’re choosing between them, compare the quality of an approved output and the time it takes your team to check it, as well as the model’s response time.

Source · OpenAI


A purple prism with coloured light trails.

AI models

3. Google introduces Gemini 4 Neon

Google announced Gemini 4 Neon on 18 November, positioning it for faster completion of complex coding, research and knowledge-work tasks. Google’s launch results are a starting point. A sensible business trial would compare Neon against the model you already use on the same inputs and check whether the outputs meet your criteria.

Sources · Google · Google DeepMind model card


A calendar next to a green checkmark.

Workplace automation

4. Microsoft begins rolling out Copilot Autopilot

Autopilot is designed to keep track of recurring work, follow up on threads and return to a project over time. Microsoft began rolling it out to eligible Frontier customers on 17 November. A useful first test would be a weekly project summary based on one approved Teams channel, with a project lead checking the draft.

Sources · Microsoft Frontier · Microsoft Ignite


A laptop with overlapping windows and a cursor.

Computer use

5. Gemini can operate Mac apps in the background

Google’s Gemini app can now click, type and interact with approved apps and websites in a separate background window, with a live preview for the user. The initial computer-use rollout is in English for eligible US personal accounts on supported Macs. For New Zealand teams, the workflows are worth watching while local availability develops.

Source · Google


An illustrated processor with gold traces across a red background.

AI infrastructure

6. Chinese technology firms face uncertainty over AI chips

Associated Press reported that Alibaba and Tencent were bracing for potential AI development setbacks amid tighter US restrictions on advanced chips. It’s a reminder that model and cloud availability depend partly on the hardware and regions a provider can serve. Teams planning a major integration should know which regions and providers their setup depends on.

Source · Associated Press


A folded newspaper next to a chat bubble.

Digital publishing

7. Le Figaro strikes a distribution deal with OpenAI

The French publisher and OpenAI announced an agreement for Le Figaro reporting and archive material to be discoverable in ChatGPT through linked summaries and excerpts. For publishers, the development raises a practical question: which AI channels send readers through to the original story, and how will those visits be measured?

Sources · Le Figaro · OpenAI


A browser window beside a testing flask.

Websites

8. WordPress 7.2 reaches release candidate

WordPress released the first 7.2 release candidate on 17 November, with the final release scheduled for 8 December. The candidate includes editor and collaboration improvements. For site owners, this is a chance to test the combination of theme, page builder, forms and plugins on staging before planning a live upgrade.

Source · WordPress.org


A cloud behind a digital storefront.

Website automation

9. Cloudways opens its AI website builder to general use

Cloudways announced general availability of a tool that turns a written brief into a WordPress or WooCommerce site that users can customise and host on its platform. That can shorten the first build. Businesses still need to test the customer journey, including enquiry forms, mobile navigation and, for stores, checkout.

Sources · Cloudways · DigitalOcean


A painter’s palette alongside a switch in the off position.

Design tools

10. Canva adds controls for generative AI

Canva introduced a switch for generative AI features and a separate option to see less AI-generated material in its media library, available on Canva Free and Pro. The distinction matters for teams that want to use familiar editing tools while setting boundaries around generated designs or assets.

Sources · Canva · The Verge


Editorial illustrations generated for ADJ Systems. This preview edition reflects reporting available on Saturday afternoon, New Zealand time.


A cursor selects a spreadsheet cell alongside other app windows on a laptop.

Google Gemini can now use Mac apps. What does that mean for businesses?

Google has added computer use to Gemini for Mac. Give it a multi-step task and it can click, type and work across approved apps in a background window while you carry on using your Mac.

For businesses, that points to a new way of handling repetitive work that crosses several systems. The immediate question is which steps are predictable enough to delegate and where a person should check the result.

Read Google’s 19 November announcement

Availability at launch

US only, in English. Eligible Google AI Plus, Pro or Ultra subscribers aged 18+ using a personal Google Account and an Apple Silicon Mac with macOS Sequoia 15 or later.

Google’s eligibility and setup guide


Header image: AI-generated editorial illustration for ADJ Systems.


See Gemini work across an app

In Google’s launch demo, Gemini takes a child’s butterfly sketch and builds and rigs a 3D model in Blender. The demo shows how Google wants the feature to work inside a desktop app. Your own workflow would need its own test.

Video: Google. Gemini models and rigs a butterfly in Blender. Watch on Google’s announcement page.


How computer use changes the workflow

A chatbot can tell you how to complete a task. With computer use, Gemini can also take the steps in the apps you’ve allowed. Google puts those actions in a background window and shows a live preview in the chat. You can stop the task and confirm sensitive actions when prompted.

An example with a human review step

Imagine processing a receipt across a downloads folder and an expense tool. Here’s how a small pilot could divide the work.


1. Define the task

Ask the agent to extract the supplier, date and amount and prepare an expense entry.

2. Watch it work

Give it access to the folder and expense tool. Follow the live preview as it fills the fields.

3. Check and submit

Compare the entry with the receipt. A person corrects any errors and submits it.


Workflow diagram: ADJ Systems. Setup and controls are described in Gemini Apps Help.

Where businesses could use it

Look for work that repeatedly moves information between approved tools. These are examples of workflows a business could test when suitable tools are available.


Customer enquiries

Group incoming requests, flag missing details and prepare a CRM update and draft response for a staff member to check.

Management reporting

Collect figures from approved dashboards and build a draft weekly report. Check the figures against the source before circulation.

Content operations

Move approved copy and media credits into a CMS draft. Keep layout review and publication with the editor.


What still needs a person

When an agent enters data or changes a record, a wrong field or recipient can affect customers and colleagues. Start with sample data, restrict access to the apps the task needs and keep approvals in place for sending messages, updating live records or making payments. A live preview helps you follow the work. Check the final result against the source before it goes further.

What New Zealand businesses can do now

As of this test article, Google lists the United States for this rollout. In the meantime, choose one recurring task, write down its inputs and approval points, and time how long it takes today. Include exception handling and checking time in that baseline.

If your tools already have a reliable integration, use it for predictable transfers. An agent may help with steps that still need someone to navigate an interface. A useful pilot will show whether the whole task takes less staff time once review is included.

Sources


A product catalogue and headphones sit opposite a chat bubble and a shopping basket.

Shopify’s AI storefronts: What New Zealand retailers need to know

Shoppers can increasingly find products inside AI conversations. Shopify’s Agentic Storefronts give eligible merchants a way to make their products available to AI channels, with checkout options that depend on the channel and the shopper’s market.

For a New Zealand retailer, the starting point is the catalogue. An assistant can only accurately recommend a product for a particular size, delivery deadline or budget when it has dependable information to work with.

Read Shopify’s Agentic Storefronts guide

What’s available

Shopify says Google AI Mode and Gemini are rolling out for eligible stores selling to customers in the US, Canada or Australia. A store can be based in another country. Direct checkout is rolling out separately. See the channel requirements.


Header image: AI-generated editorial illustration for ADJ Systems.


See an AI shopping journey

Shopify’s example follows a shopper looking for a lamp in a chat. It shows the handoff from a product recommendation to checkout. The available shopping and checkout options depend on the channel.


Discovery, checkout and fulfilment are different steps

A shopper might ask for over-ear headphones that fold flat, come with a case and can reach Sydney before an upcoming trip. An assistant has to match the request to product and delivery data. Depending on the channel, the shopper may complete payment in the conversation or be sent to the retailer’s site. The retailer still needs to fulfil the order and handle support.


1. Supply useful catalogue data

Use meaningful product titles and specific attributes. Put dimensions, compatibility, variants and other facts in structured fields where supported.

2. Help shoppers choose

Keep availability and prices current and describe when a product fits a particular use. Make important conditions visible in the listing.

3. Test the handoff

For an eligible market, follow the route from discovery to product page or checkout. Check currency, delivery options and the return information shoppers see.


What changes for a New Zealand retailer

A store that only sells to New Zealand should check each channel separately. Shopify’s published Google AI Mode and Gemini criteria require selling into at least one of the listed customer markets. For retailers that already serve Australia, this makes Australian catalogue and shipping data a sensible place to start. Check the Agentic sales channel in Shopify admin for the store’s status and use the relevant channel guide before making rollout plans.

A catalogue audit you can do now

Choose 10 products and write realistic questions a shopper might ask. Check whether the catalogue provides an accurate answer for each one, including variant, price, dimensions and delivery destination. Compare those answers with the product page and checkout. Fix inconsistencies at the source so the same update can benefit search, your own storefront and any eligible AI channel.

When a new channel becomes available, judge it on qualified traffic, successful orders and avoidable support questions. Start in one market so you can distinguish catalogue problems from rollout differences.

Sources