Why Shopify's AI Assistant could transform UX, NotebookLM's new superpowers, Microsoft's "agentic web"
Plus: Is cheating in interviews fair? How to design buttons, Anthropic takes us one step closer to replacing traditional software engineers.
Hi product people 👋, Rich here with the weekly briefing.
To say it’s been a big week for tech would be quite the understatement with major new updates from Google, Microsoft, Shopify and more. I’ve done my best to wade through hundreds of announcements to try and figure out which ones matter most to product teams.
We’ll explore Shopify’s new AI Assistant and why its new UX pattern could transform onboarding and troubleshooting journeys in SaaS products, Google’s impressive new AI features in Workspace and how Microsoft’s leadership team view the future of the so-called “agentic web”.
Plus, the dramatic and destructive impact AI is having on SEO and why one startup founder thinks the only fair way to do interviews in 2025 is to let everyone cheat.
Happy Friday and enjoy your weekend!
More: Watch on YouTube / Follow on Substack Notes / Unlock the Full Newsletter
Shopify, Google and Microsoft reveal their next big things
First up, after their CEO declared AI as a non-negotiable part of life at Shopify, it’s pretty clear that AI is now firmly centerstage of the company’s product strategy. This week, Shopify unveiled over 150+ major new features as part of the company’s biannual Editions.
One of the most impactful releases for product teams is the set of new features and capabilities added to Sidekick, Shopify's AI Assistant. It can now run reports with enhanced reasoning capabilities and generate visual assets that can be used in stores. And one feature in particular could be transformative: it now has new voice chat capabilities that let users share their screen to co-browse Shopify's dashboard together. Users can "call" up the assistant and ask it for help with troubleshooting or understanding how the product works. This looks pretty impressive and could mark a shift in UX for how users work with AI assistants in SaaS products - especially for tasks like onboarding.
Other updates include:
Storefront MCP - a new MCP server that lets product teams build AI agents that can access Shopify’s product catalogue.
A generative AI theme builder for Shopify’s core theme, Horizon that lets users describe the assets they want and the theme builder handles the rest.*
Meanwhile…
Google unveiled some seriously impressive new AI capabilities during its I/O event. There are far too many to mention here but for product teams who use Workspace, some of the relevant updates to check out include customizable avatars that can present Google Slides presentations for you, real time speech translation in Google Meet and a new tool called Stitch which is Google’s attempt at a “vibe coding” / design tool. Stitch is free to use right now and is still very much in its beta stage but worth a look nonetheless.
The company’s new video models are also extremely impressive. Veo 3 comes with the ability to generate videos with sound and it’s now getting much harder to tell the difference between AI generated content and non-AI content.
Google says its new video models will be rolled out across its portfolio of products - including NotebookLM which will get video generation capabilities alongside audio. See it in action here.
And while we’re on that topic, Klarna’s CEO also dabbled with the use of AI avatars in the company’s latest earnings call. It feels a little bizarre for a CEO to do this but it does demonstrate a willingness to embrace AI in a way that’s a little less controversial than the company’s previous bad press around ditching humans for AI in customer service.
GitHub’s new coding agent is finally unleashed - here’s what developers think…
Microsoft hosted a major event of its own where its leadership team shared their vision of what the Agentic Web might look like.
One of these agents is GitHub’s new Coding Agent. This latest iteration of Copilot means it can act as an autonomous coding agent, capable of handling tasks like bug fixes, adding new features, and code cleanup. Users can assign tasks directly to Copilot and it will pick the ticket and let you know when it’s done. Human reviews of the code are still mandatory at this stage and early hands-on reviews from engineers are pretty positive. In this example, GitHub’s agent was assigned a task and the pull request was, according to the engineer, “spot on”.
Microsoft also announced a new research and development platform called Discovery that uses AI agents to tackle complex scientific research. Microsoft says the platform has already aided in the development of a new coolant for data centres, cutting down a process that would have taken months or years to 200 hours.
OpenAI previews its own new AI agent Codex
Get up to speed with OpenAI’s new Codex Agent: what it does and how companies (and product managers) are using it.
*Includes sponsored content
Key reads and resources for product teams
New from the Department of Product Substack this week:
Knowledge Series - Multi AI agent architecture explained
In this Knowledge Series, We’ll look at the most common types of multi agent architectural patterns along with some real world examples from top companies to bring everything to life. If you’re curious about how AI agents work, and in particular how they can work together to achieve common goals, then this should help.
Deep - The AI companies that want to replace us
Plus, what it might take to become irreplaceable and thrive in the AI-disrupted workplace in 2025 with a curated selection of insights and opinions from top tech CEOs from companies including Figma, Nvidia, Microsoft and more.
More reads to feed your product brain
How to design button states correctly (NN Group)
Microsoft’s CTO Kevin Scott on how AI can save the web, not destroy it (The Verge)
Understanding the 5 types of AI consumers (Bain Consulting)
How generative AI could disrupt SaaS (Harvard Business Review)
Tools you can use
Prism - AI that watches your session replays, catches bugs and suggests optimizations.
BnbIcons - generate your own icons inspired by Airbnb’s new design system.
Propolis - Full QA coverage, zero setup. Propolis simulates real users via swarms of autonomous browser agents that test for you.
📈 Product data and trends to stay informed
Google’s AI Overviews are leading to a 56.1% lower CTR on desktop and 48.2% CTR lower on mobile for the newspaper publisher Mail Online.
The company’s director of SEO described the drop as “pretty shocking”. 25 to 34 year olds on mobile are the power users of AI overview, picking AI 50% of the time and when users do click, they choose brands that are recognised and trusted like Reddit and Youtube.
Stack Overflow has been particularly badly hit by AI, with traffic plummeting.
69% of UK streamers now say they prefer watching free ad-supported content and 20% plan to add a FAST service in the next 12 months. Interesting report from LG that could influence monetization strategies for other tech companies.
Anthropic’s new Claude 4 model has achieved a 72.5% score on SWE-bench - the software engineering test that assesses how well AI can complete engineering projects. This beats OpenAI’s score of 69.1% and Gemini’s 63.2%.
Entry level tech jobs are evaporating for new graduates, with many blaming AI. The unemployment rate of recent graduates has jumped from 4.6% to 5.8%. Gen-Z is turning to alternative career paths as a result. The proportion of students taking vocational degrees is up from 15% to 20%.
AI job interviewers are going viral on TikTok - for all the wrong reasons. Here’s an example.
The number of people with brain-computer interfaces is set to more than double in the next 12 months.
Other product news in brief
🤖Apple has delayed the rollout of AI Siri. Insiders say the failure to get AI right threatens iPhones dominance and future plans for robotics.
💡OpenAI is acquiring Jony Ive’s design consultancy firm and Ive is officially working with the company on potential new devices. Ive says that Rabbit and Humane were both bad products and that the new product won’t be a wearable which makes me more optimistic about whatever Ive and Altman are cooking up.
🧐A startup founder thinks the only fair way to do interviews in the AI age is to allow everyone to cheat. Here’s why.
Paid subscribers get the full DoP Substack including: The Knowledge Series for sharpening their tech skills, AI tutorials for putting AI into practice at work and DoP Deep dive reports to learn from the world’s top tech companies.