Google Sheets gets AI functions and Cursor lets you delegate work to agents on your mobile
Plus: Is OpenAI building a rival to Microsoft? Head of ChatGPT on the 3 essential skills he looks for, why you shouldn’t use AI for forecasting, is pay per crawl the best model for the AI web?
Hi product people 👋,
Coming up this week, Google Sheets officially gets support for a new AI function and Cursor releases a new web and mobile app that lets product teams assign work to it and get pinged when it’s ready for review.
We also take a look at the latest product-related nuggets from the impressive State of AI report and while it may be independence day in the US, one CEO has decided to rename today to push for a new monetization model for the web in the AI age.
Happy Friday and enjoy your weekend!
Rich at DoP
More: Watch on YouTube / Follow on Substack Notes / Unlock the Full Newsletter
New product features and innovation this week
First up this week, Google Sheets is getting a brand new “AI function”. The new function is initiated using “=AI” and right now it can be used to generate text, summarize information, categorize information and analyze sentiment. I’m surprised it’s taken this long given just how many third party apps there are that do this, but it’s great to see native support added nonetheless.
But Google Workspace has got itself some new competition
After acquiring Coda a few months back, Grammarly has acquired email startup Superhuman. It’s clearly attempting to join Notion, Google Workspace, Microsoft and others in building an AI-powered set of productivity tools. Its ultimate goal is to become an “AI productivity platform”. Grammarly recently raised $1 billion to fund the spree, with proceeds of future revenue being returned back to investors.
But as well as Grammarly, Google may also find itself competing against OpenAI. Reports this week suggest they are quietly testing a rival set of products to Office and Workspace. This apparently includes collaborative document editors with integrated chat capabilities and a new browser - pitting it directly against Google and Microsoft. If this is true, it’s increasingly becoming easier to see why the tension is mounting between OpenAI and Microsoft.
Meanwhile…
Cursor has released a new web and mobile app. It lets you type in a task and an AI coding agent will get to work on the task in the background. When the agent completes the work, developers can carry on the agent's work directly in the code editor in Cursor.
Anthropic’s Claude has released a new way to use the “USB-C of AI” applications. Claude Extensions lets users install MCP servers in one click. On launch, there are MCP servers available from Chrome, Apple Notes and iMessage - as well as a local file system MCP server that lets Claude edit your local files. Companies are now racing to adopt MCP internally with Amazon leading the way. The Pragmatic Engineer says that Amazon is likely now the global leader in MCP with one engineer saying that “most internal tools and websites already added MCP support”.
Key reads and resources for product teams
New from the Department of Product Substack this week:
Knowledge Series - How to set up your own MCP server - a practical guide for non-engineers
Get up to speed with hands-on experience of using MCP. In this Knowledge Series, we’ll cut through the noise and focus on the practical steps to set up your own MCP server. This guide walks you through both the manual process and Claude’s new method, offering real-world examples to help you understand and apply MCP servers in your daily work. (Department of Product)
UX - How to design prompt suggestions for AI tools
Use-case prompt suggestions vary widely in complexity — from short phrases or single-sentence examples to rich, realistic formats like entire conversations, images, or videos that show full user–system interactions. (NN Group)
More reads:
How to build “word of sight” into your product (Marketing Ideas)
A new study shows the downsides of using generative AI for forecasting (Harvard Business Review)
OpenAI’s Head of ChatGPT on the future of Programming and the 3 essential skills for today (OpenAI)
Tools you can use
Tidbit - transform ChatGPT into a Slack-type user interface with channels and multiple users.
OpenRouter - automatically selects the best AI models for your product. Product teams need to integrate into a single API to get access to multiple AI models. Raised $40 million in funding last week.
Lazy - capture and manage all of your thoughts in one place. Lets you capture and collect from multiple sources including web pages, Kindle books, YouTube videos, social media and more.
📈 Product data and trends to stay informed
Describing new features as “AI powered” doesn’t make users more likely to buy your product. A new study showed that for 58% of users, the addition of AI powered features made no difference at all to their purchase decision. For almost a quarter of users, it made them less likely to buy.
To be fair, the study includes smart home appliances which users are notoriously sceptical of (who needs an AI powered toothbrush?), but the data is interesting nonetheless.
Banks have started to roll out “digital employee personas”. The two main digital employee personas developed by BNY Mellon are the Code Vulnerability Cleanup Persona Designed to identify and fix vulnerabilities in code and a Payment Instruction Validation Person which Focuses on validating payment instructions, ensuring that transactions are processed correctly and securely.
Bots now outnumber humans on Zoom calls in some companies. One call had 6 humans and 10 AI note takers.
OpenAI’s head of API product says that API usage is up 700% year over year and over a million monthly active developers are now using it. As agent complexity grows, companies are shifting from single-agent models to modular sub-agent architectures, where specialized agents handle different tasks. He says Stripe has reduced its invoice resolution time by 35% thanks to AI agents.
This reliance on third party APIs is echoed in a new State of AI Report 2025 which says 80% of companies use third party APIs to power their AI features.
Other interesting nuggets from the report:
30-45% of product roadmaps are now dedicated to building AI-driven features. In high growth companies, 43% of roadmaps are focused on AI - up from 31% in 2024.
79% of AI-native and 65% of AI-enabled companies are building agentic workflows
40% include AI features as part of a premium-tier product; 33% include them at no extra cost.
And finally…
Is pay per crawl the best business model for the AI web?
According to Cloudflare, with OpenAI, it's 750 times more difficult to get traffic than it was with the Google of old. With Anthropic, it's 30,000 times more difficult. CloudFlare’s CEO argues that the pay per crawl model is a better way to reward companies who miss out on this traffic. And to celebrate the launch of this new feature, he’s declaring today content independence day.
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 world’s top tech companies.
The problem with all AI those who can afford it can use it. ChatGPT only give you so much, I wanted information which it was converting to pdf and allowed me to download, but only twice, what good is that, its the same with every AI which offers free, you only get the badic unless you can afford it. Its about time there was a company that grew some balls and gave the world full function AI, even running adverts every 15 minutes. AI is going to be the bases for every childs education in the future, but only if the household can afford it, then it will be education for the wealthy, we see it in top flight Universitys next will be all education for rich in every school, the poor gets lower than basic all because greed of the tech companies keeping AI for those who can afford it.