Claude comes to Chrome and the 5 AI prompts Microsoft's CEO uses every day
Plus: The latest Top 100 Consumer AI apps report, Google takes on DuoLingo, a new YC tool to run rapid audience experiments and test new product concepts
Hi product people đ,
Coming up in this weekâs briefing, a new extension from Anthropic sees Claude join the AI browser wars, a new tool that will let you test product features and messaging on AI âArtificial Societiesâ and Microsoftâs CEO shares the 5 prompts he uses every day.
Happy Sunday and enjoy the rest of your weekend!
Rich
Watch on YouTube | Follow me on Substack Notes
Anthropic bets on a new extension - but security risks remain
First up, Anthropic has released a preview of Claude for Chrome - a new Chrome extension where users can instruct Claude to take actions on their behalf in the browser. Anthropic says that during its early beta tests with 1,000 users, so far people are using the extension for tasks like managing calendars, scheduling meetings, drafting email responses and one use case that could be particularly interesting for product teams: testing new website features.
But, Anthropic says it is focusing heavily on security before a wider rollout, citing one example of a case where a malicious email encouraged the Claude browser agent to delete a bunch of emails - which the agent did without the userâs permission. Brave has also previously shown how Perplexityâs Comet browser is vulnerable to prompt injections. Anthropic clearly sees the browser as a key battleground for practical AI use cases but unlike Perplexity, Microsoft, Google and OpenAI, it looks as though it wonât be attempting to build its own AI-powered browser.
Perplexity shipped some simple but friction-reducing new UI updates to its mobile app including new animations for voice interactions, a new default state when you open the app that has the keyboard open by default and the ability to swipe between search and discovery. Perplexityâs head of design Gunnar Gray described the changes as âa more refined and focused experienceâ. After updating to the latest version of the app, that seems like a pretty good description of the refinements.
Elsewhere, Spotify is continuing its shift towards becoming a social network by launching a new messaging feature which lets users send message requests to other Spotify users. If they accept, they can use the DM feature in much the same way as other social networks. Itâs been almost 5 years since Spotify first announced high fidelity audio with no release date as yetâŚ
Google takes on DuoLingo - and ships a bunch of new AI features
Itâs been widely noted that DuoLingo is now using AI to generate the bulk of its language learning lesson content. And it looks as though Google took notice of this. This week, Google has unveiled a new feature in Google Translate that will not only translate content - but also teach you languages. Powered by Gemini, the new feature will create lessons based on your skill level and purpose for learning a language such as vacationing in a different country. The new Translate app also comes with new live translation capabilities.
Google has also revealed an impressive update to its image editing model with Gemini 2.5 Flash- also known as nano-banana. The model produces higher quality images than previous models, allows you to blend multiple images into a single image and integrates Geminiâs knowledge of the world. Google says this means it can understand simple hand drawings and make edits of those which could be helpful for things like sketching prototypes. Hereâs some samples of it in action.
And Vids is getting lifelike avatars that could be super helpful for things like product demos. Users can type the script they want to use and Vids will generate a lifelike avatar of the content.
Microsoft consciously de-coupling from OpenAI?
Microsoft is launching its own in-house AI models for the first time. The models are called MAI-Voice-1 and MAI-1-preview. The company says its new MAI-Voice-1 speech model can generate a minuteâs worth of audio in under one second, while MAI-1-preview âoffers a glimpse of future offerings inside Copilot.â You can try MA1-Voice-1 out for yourself on Copilot Labs, where you can enter what you want the AI model to say, as well as change its voice and style of speaking. Hereâs a sample of it in action.
Microsoft says:
We have big ambitions for where we go next. Not only will we pursue further advances here, but we believe that orchestrating a range of specialized models serving different user intents and use cases will unlock immense value. There will be a lot more to come from this team on both fronts in the near future. Weâre excited by the work ahead as we aim to deliver leading models and put them into the hands of people globally.
Clearly, the fractured relationship between Microsoft and OpenAI has worsened and we could see Microsoft decouple itself entirely from OpenAIâs models in the future if these proprietary models perform well enough. Still, for the moment at least, itâs currently in negotiations to extend their partnership well into 2030.
Key reads and resources for product teams
New from the Department of Product Substack this week:
Deep - How are companies driving engagement and retention?
Real world examples of new engagement and retention tactics from companies including Figma, Reddit, Instacart, Replit, Instagram and more.
Knowledge Series - How to use Claude Code for non-engineering use cases
Claude Code is fast becoming the go-to AI tool for product teams - and it isnât just for engineers. In this Knowledge Series, discover how to use Claude Code for creating agents for PRDs, performing SEO audits, building a second brain and more. (Department of Product)
A designerâs guide to AI-powered prototyping - how to use AI tools to explore design ideas in high fidelity (Adobe Design)
How to refactor an entire codebase with Cursor - an in-depth demo of how engineers might refactor a codebase using AI (YouTube)
5 prompts Microsoft CEO Satya Nadella is using every day - prompt templates you can use in GPT-5 for meeting prep, tracking KPIs, measuring product progress and more. (Satya Nadella)
Tools you can use
Artificial Societies - run rapid audience experiments and test new product concepts with Artificial Societies. A Y Combinator startup that this week raised $5.3 million in seed funding.
Pylon - new competitor to the likes of Intercom for B2B customer support. Pylon consolidates multiple different support channels from Slack, Teams and others into one place. Automatically identifies knowledge gaps, generates help articles from customer conversations.
AI thing - a handy little tool that can handle complex or repetitive tasks in the background. It lets you use your own agents across tools like Notion, GitHub and Google Workspace - with plans to add more.
đ Product data and trends to stay informed
Anthropicâs latest Threat Intelligence Report revealed that North Korean operatives have been systematically leveraging Claude to secure and maintain fraudulent remote employment jobs at tech companies.
61% of AI in Claude activity was used for front end development, 26% for scripting, 10% for interview prep and 3% for backend development. The report also outlines other case studies including an AI powered romance scam bot that has over 10,000 monthly users and instances of âvibe hackingâ - where AI agents are deployed at scale. Read the report in full.
This week, the latest Top 100 Consumer AI apps report was released from Andreessen Horowitz. Some key stats from the report worth knowing:
ChatGPT still leads on mobile, and Gemini is in second, with nearly half as many monthly active users (MAUs) as ChatGPT.
NotebookLM is slowly declining after peaking in April - this may be seasonal, letâs see.
Grokâs companion feature seems to be driving DAU engagement - if so, others may follow suit but itâs unlikely theyâll be as bombastic / controversial as Grokâs (which may be a driver of retention).
Vibe coding site traffic is also down but retention is looking fairly strong.
Junior software developers are struggling. A new piece of analysis in the Wall Street Journal shows that Among software developers aged 22 to 25, the head count was nearly 20% lower in July 2025 compared to its late 2022 peak. Now to be fair, this report uses October 2022 as its 100 index which was the peak of the overhiring craze. The data doesnât show junior hiring levels before 2021 which could well be below the current rate.
The shadow AI economy is alive and well. In MITâs State of AI in Business Report, it was found that while only 40% of companies have purchased an official LLM subscription, 90% of employees said they use LLMs regularly.
Payments company Block says its AI agent framework, Goose, is now used by 4,000 engineers, with adoption doubling monthly. Goose writes about 90% of code for its users and has saved engineers an estimated 10 hours of work per week.
Taco Bellâs AI voice-powered drive throughs have now processed more than 2 million orders so far. But Customers have reported glitches, delays, and some have intentionally trolled the AI with absurd orders like âlike, â18,000 cups of water, pleaseâ. Their CTO said âweâre learning a lot, Iâm going to be honest with youâ.
A âback to schoolâ reading list
As summer draws to a close, hereâs a selection of books recommended by leading faculty members at Harvard Business School:
Paid subscribers get the full DoP Substack including: The Knowledge Series for sharpening your tech skills, AI tutorials for putting AI into practice at work and DoP Deep dive reports to learn from the worldâs top tech companies.