🔵 Google Stitch gets Skills and Stripe's CEO says this is the "coolest breakthrough since GPT-4"

Plus: How to use Claude Code with Notion, Satya Nadella on the future of product teams and Anthropic leads the LLM API market - but for how long?

Hi product people 👋,

This week saw global tech leaders head to Davos to outline their views on what the future looks like and among the lofty predictions about the societal impact of AI, there were some interesting, practical insights from Microsoft’s CEO Satya Nadella about how the company has reconfigured its product teams.

Google also held its first Stitch Developer Week with some major announcements for product teams including new MCP server support, Agent Skills for design systems and fresh documentation on how to actually prompt with Stitch effectively.

Plus, new leaked data showing Gemini API calls more than doubled to 85 billion (and why Anthropic now dominates LLM market share), fresh analysis debunking the “search is dying” narrative, and a warning that vertical SaaS markets once considered “AI-proof” are now under threat.

As always, feel free to drop any comments about anything featured this week below!

Rich

Watch on YouTube | Follow on Substack Notes

---

Key reads and resources for product teams

New from the Department of Product Substack this week:

Knowledge Series - Agent Driven User Interfaces Explained

Google’s former CEO Eric Schmidt has predicted that in the future, user interfaces as we know them “are going to go away”. In this Knowledge Series, we’re going to unpack the essentials of A2UI that product teams need to know including the core technical pieces that power the technology, real world examples in action plus how you can get some hands on experience with A2UI to get up to speed quickly.

New in the AI library - Screenshot feature extractor Skill

This Skill uses a multi-agent analysis system with 5 different AI Agents to extract product features and analyze them across important dimensions. When you ask Claude to analyze competitor UIs, turn design mockups into requirements, or identify features from app screenshots, this Skill coordinates five specialized agents that examine UI components, user interactions, and business logic in parallel, then synthesize everything into an actionable markdown checklist of what needs to be built - without prescribing how to build it technically.

(Department of Product)

Report - The State of UX 2026

UX is stabilizing, but the rules have changed. In this report, NNGroup unpacks why UI alone won’t differentiate your product in 2026, how AI is shifting from hype to fatigue, and what “judgment over artifacts” really means for product teams. (NN Group)

Tools you can use - Kimi Slides

This is an alternative tool for creating presentation slide decks at work. Some people have described it as the “NotebookLM for slide decks” and “the biggest underhyped thing right now”. The demos of it certainly look pretty impressive. Worth a look if you’re interested in exploring alternative tools for creating presentation decks. (Kimi)

Analysis - Debunking the myth that Search is dying

Is AI really eating into traditional search? A new paper challenges the narrative that “SEO is dead” by analyzing traffic data from over 40,000 websites. While many studies claim search traffic has plummeted 25-50% due to ChatGPT and AI tools, the data shows organic search is down only 2.5% year-over-year. Traffic to Google actually increased slightly in 2025, and 90% of clicks still go to organic results rather than ads. (Graphite)

Process - How to use Claude Code with Notion to oversee the work of AI coding Agents

Notion’s design engineer, Geoffrey Litt shares how he visualizes the work of Claude Coding agents with Notion through live Kanban boards. (X)

The future of product teams - Satya Nadella on how LinkedIn has changed the composition of its product teams

Satya Nadella says AI tools are doing for product development what PCs and Excel did for forecasting - eliminating the friction between different specialist roles. He says that at LinkedIn, AI capabilities now allow one person to handle what previously required handoffs between PM, design, and engineering roles. (YouTube)

---

YouTube’s product roadmap for 2026 is ambitious but contradictory

YouTube’s CEO Neal Mohan has outlined the company’s product roadmap for 2026 - and it’s bold but a little contradictory. On one hand, they’re giving creators even more AI tools including the ability to create an AI avatar clone that can be used on Shorts, and on the other, they’re promising to crack down on AI slop with a commitment to building new AI slop detection systems.

There’s also a renewed focus on both Shorts which Mohan says averages 200 billion daily views and on TV (where YouTube is now leading). The one group that seems to be increasingly forgotten by YouTube’s product leadership team, though, is the average desktop user who watches longer form videos. A quick glance at the YouTube homescreen shows that the grid of longer form videos has shrunk with more prominence given to Shorts. There’s a risk that by trying to cater to each of these different viewing habits, they move too far away from (what was once at least) their core value proposition.

TikTok has also extended its formats into long form and this week, Netflix unveiled its product roadmap for 2026 with an announcement of plans to overhaul its mobile app. The redesigned app is set to focus more on short form videos and podcasts and their CEO said the revamp was designed to “better serve the expansion of our business over the decade to come”.

Stripe’s CEO on the “coolest breakthrough since GPT-4”

Stripe’s CEO says this is the “coolest breakthrough since GPT-4”. The tl;dr is that Cursor figured out how to run hundreds of AI coding agents autonomously for weeks on massive projects, writing over 1 million lines of code. In one experiment, they asked the coding agents to build a browser and in another they asked it to clone Excel. Read the paper here.

Cursor has also released an internal tool called BugBot to the public. Bugbot is a code review agent that analyzes pull requests to catch issues before they reach production. Cursor says that the system has a resolution rate over 70%, meaning most bugs it flags actually get fixed by developers. It currently finds about 0.5 resolved bugs per PR on average. The full write up on how they built it is worth a read if you’re interested in reducing the number of bugs shipped to production.

Google Stitch gets more powerful with MCP and Skills

Google Stitch held a Developer week this week which saw some major announcements for product teams. If you’ve not used Stitch before, it’s the Google equivalent of Lovable or Figma Make that works by first creating design artifacts in a canvas and then “stitching” them together.

The announcements included a new MCP server which allows AI coding agents to interact directly with Stitch designs and new Agent Skills. The Agent Skills include a design-md Skill which generates a robust document outlining your product’s design system that Google says can act as your design “source of truth” and a react-components Skill which converts your designs into React components. You can grab the Stitch Agent Skills here.

As well as these announcements, Google also published a new set of documentation that not only includes technical set up guides for developers but also some pretty handy how-tos for prompting in Stitch.

How to prompt effectively with Stitch (according to Google)

An example of Google Stitch in action - building a plant care guide for my too often neglected plants

In its Stitch Docs, Google recommends being specific and iterative when designing apps with Stitch. Here’s some of the prompting techniques you can use from its guide to get the best results:

• Starting your project - begin with either a broad concept (”An app for marathon runners”) or get detailed upfront by describing core features. Use adjectives to set the visual tone since “vibrant and encouraging” will create very different designs than “minimalist and focused”

• Refining screen by screen - make one specific change per prompt and target exact elements. Instead of asking for sweeping changes, try something like “Change the primary button on the login screen to larger and use brand blue”

• Controlling your theme - request specific colors (”forest green”) or describe moods (”warm, inviting palette”). For fonts, specify styles like “playful sans-serif for body text” or “serif for headings”. When modifying images, be precise: “On Team page, Dr. Carter’s photo: update lab coat to black”

• Other tips - use UI terminology (navigation bar, hero section, call-to-action), iterate gradually rather than asking for multiple changes at once, and reference brand elements when making adjustments to maintain consistency

Spotify’s Page Match is a delighter and a new experiment in product growth

Spotify is testing out a new feature called Page Match that will allow users to scan a page of a physical book and start listening to an audiobook from that page. Spotify previously acquired an audiobook company back in 2021 and given the low margins on music, audiobooks are a potentially much better revenue stream for the company.

While this feature is a “delighter” in many ways, it’s also a discovery and activation tactic to try to drum up some of that revenue; users don’t always know what’s available in the Spotify library and so this new feature removes any friction by creating a new entry point. If users already own a physical version of a book, why not try the audio version? The question for Spotify, though, is how many users who own the physical version of a book also want to try out the audio version. Nonetheless, it’s a pretty creative attempt at product growth.

---

📈 Product data and trends to stay informed

New leaked data shows that API calls to Gemini models more than doubled from around ~35 billion in March 2025 to ~85 billion in August 2025:

Anthropic now dominates LLM API market share by usage and Google looks set to overtake OpenAI very soon, too. Its market share increased from just 12% in 2023 to 40% in 2025. When your product is an API, switching costs are less relevant - and it’s for this reason that Google could eventually also replace Anthropic. Alphabet’s upcoming earnings will give us a closer insight into Gemini’s revenue streams.

Publishers expect traffic from search engines to decline by 40% over the next 3 years according to new analysis published by Oxford University.

The gap between the impact of AI on productivity between senior execs and workers is huge. 40% of non-exec staff say they saved no time with AI vs just 2% of C-suite execs.

Vertical SaaS is under threat. A new RBC Capital Markets report warns that AI is now penetrating vertical SaaS markets once considered “AI-proof” due to regulatory complexity and domain expertise requirements. Following Anthropic’s launch of three products this past week - including Claude for Healthcare & Life Sciences, Claude Cowork and an expanded Labs incubator - shares of some major software companies like Salesforce, Workday, Intuit, and Snowflake fell 6-13%. The report suggests this “AI overhang” on software stocks will likely persist and spread throughout 2026. For product teams, the critical strategic question is increasingly becoming: what can you own that an AI model can’t replicate?

Independent analysis from SensorTower shows that Amazon’s Rufus AI Assistant drove 3.5× higher conversion than non-Rufus sessions.

---

Paid subscribers get the full DoP Substack including: The Knowledge Series for sharpening your tech / AI skills, the AI Prompt and Skills library and DoP Deep dive reports for in-depth analysis to learn lessons from the world’s top tech companies.