🔵 Google Search now generates UI on demand

Will this set a new benchmark for in-product search capabilities? Plus: takeaways from Anthropic's Code with Claude event, Figma fights back and new data reveals the state of AI in design

Hi product people 👋,

This week, Google held its annual I/O event and Anthropic continued its world tour with its Code with Claude event in London. We’ll take a look at some highlights for product teams.

Plus, how DuoLingo built an internal Slack app that is now used by 30% of the company and new data from the state of AI in design reveals how the role of design is fundamentally changing in 2026.

Happy Friday!

Rich

Watch on YouTube | Follow on Substack Notes

---

Key reads and resources for product teams

New from the Department of Substack this week:

Deep Dive - How Uber’s product teams built a PRD Reviewer - and how you can build your own
Uber built an AI tool that reviews PRDs before they reach leadership - catching gaps in assumptions, blind spots, and second-order effects that individual PMs miss. This Deep Dive walks through exactly what they built, how to create your own lightweight version using Claude Projects, together with examples of similar tools from DoorDash and Atlassian that are reshaping product development processes.

New in the AI library - Strategy Tension Tool
This tool takes two inputs - your stated strategy (north-star metric + priority areas) and your actual shipped work (tickets, features, quarters) - and does the maths to show how much engineering effort actually went toward each strategic area. It then compares the two distributions, flags any area where reality diverges from strategy by more than 20%, and produces a one-page report with enough specifics to walk into a leadership conversation and say “we said X mattered, but we shipped Y” with evidence behind it. (Department of Product)

Resource - Claude Cookbooks - a selection of free practical guides on how to use Claude effectively
Claude Cookbook offers 100+ practical guides for building with Claude - from basic workflows to advanced agent patterns. (Anthropic)

UX - DesignMD tool - paste a URL and get the product’s design DNA
Paste a URL — get a production-grade DESIGN.md with colour tokens, typography, spacing scale, and component patterns. ~12 seconds. (DesignMD)

AI Org Redesign - How Coinbase built a multi-agent compliance system that puts human judgement at its core
At Coinbase, Dor von Levi and team spent six months rebuilding fraud detection from scratch, discovering that the real work isn’t checking sanctions lists - it’s the interpretive judgment in the other 95%. They built a multi-agent AI system that handles 55% of US fraud cases while adding more human review, not less, and feeds every correction back into the model. (Coinbase on X)

Interview - Notion’s CEO on the future of SaaS and how the company is reinventing itself again
Ivan explains why most SaaS playbooks are broken, why refounding might be exactly what your stuck company needs, and how organizations will actually look in three years. (Sequoia Capital)

Case study - How Spotify runs better experiments with LLMs
At Spotify, only 12% of A/B tests ship positive results - but that understates their real value. Now LLM evals are changing the game by assessing quality dimensions at scale before experiments run. Matilda Ankargren and Mårten Schultzberg explain why the relationship between evals and experiments should be a funnel, not a fork, and how this feedback loop makes both smarter over time. (Spotify Engineering)

Skills - How to get buy in for your next big idea
In this conversation, Sue Ashford, professor at the University of Michigan’s Ross School of Business, and Ellen Bailey, former vice president of business and culture transformation at Harvard Business Publishing, give suggestions for framing those ideas so that executives buy into them, including the research findings they keep in mind, questions they ask themselves and others when vetting an idea, and what they learned from the times they fell short. (Harvard Business Review)

Process - How DuoLingo built an internal Slack app that is now used by 30% of the company
Duolingo built an AI Slack app that debugs incidents, triages issues, and ships code. What started as local experiments grew into a tool used weekly by 30% of the company. Aaron Wang walks through how they evolved from painful manual setup to zero-friction automation, plus the lessons learned scaling AI agents across thousands of weekly queries. (DuoLingo Engineering)

---

New product features and innovation this week

Google’s annual I/O event took place this week and included a bunch of announcements that are worth product teams knowing about. First, Google unveiled what it is describing as its “biggest upgrade in over 25 years”. And to be fair, on first glance, it looks like a pretty accurate claim.

The “intelligent Search box”, which is Google’s first major redesign in 25 years, now uses AI to help users formulate questions with smart suggestions beyond basic autocomplete. You can input text, images, files, videos, or Chrome tabs with the UI changing depending on the user’s request.

New Search agents will monitor the web 24/7 for updates matching your criteria, automatically notifying you when relevant information appears. The new Search mode also comes with coding abilities powered by Antigravity and Google’s new model, Gemini Flash 3.5. This mode generates custom UI components on the fly - things like interactive visualizations, dashboards, and mini-apps tailored to your task, while Personal Intelligence integrates your Gmail, Photos, and Calendar data so Search understands your context.

Here’s an example of a mini application built in Search about black holes on the fly.

For product teams, this evolution of Search may not instinctively feel particularly relevant. But, if these new patterns and modes do become popular, they could set a new benchmark for a user’s basic expectations of what in-product search is capable of. More and more companies like Reddit, Amazon and others are currently experimenting with new ways to do in-product search with new trends like merging search with AI Assistants and you can check out a deep dive on that here.

Gemini Spark and new voice modes in Google Docs

Google also officially unveiled Gemini Spark - a new personal AI agent that runs on Google Cloud infrastructure rather than your device.

Spark operates as an orchestration platform within the redesigned Gemini app. You can toggle it on and assign it tasks - either through text or voice commands - and it autonomously completes them by accessing your Google workspace (Gmail, Docs, Sheets, Drive, Calendar). The agent pulls information from multiple sources, performs actions across applications, and updates your files in real-time based on incoming changes.

For product teams, you might use Spark for things like:

• Daily Brief as stand‑up prep - pair Spark’s Daily Brief–style summary with your sprint board so you get “what changed in Jira/Linear + what’s on my calendar + which docs moved” every morning.

• Prep and debrief for meetings - have Spark prep context packs (latest spec, open risks, relevant feedback) before key meetings and auto‑draft follow‑up docs and emails afterward.

• Long‑running “product questions” - start an agent task like “track all evidence related to pricing pain for Teams customers,” and let Spark collect and organize signals over weeks instead of one‑off queries.

Google also rolled out new voice capabilities, including a feature called “Docs Live” that acts as a thought partner and co-writer to help users create a first draft of a document.

But, as impressive as all of these announcements are, some folks have complained that the overall product strategy feels a little overwhelming and disjointed, with too many Gemini products, names and processes to keep track of. Despite that, the data shows that in Search at least, Google’s strategy is working - more on that later.

Highlights from the Anthropic Code with Claude event in London

Google wasn’t the only company to host an event this week, though, as Anthropic came to London to host its Code with Claude Event. I was delighted to be able to attend the event and one of the most interesting talks for me came from Spotify’s VP of engineering Niklas Gustavsson who outlined how the company is adapting to incorporate AI into its workflows.

During the talk, he shared that 99% of the company’s developers are now using AI and 73% of PRs are now AI assisted.

Code with Claude Event, May 19th, London

But, rapid adoption led to a bloated code base that was growing 7x vs the number of developers, leading to a “maintenance nightmare”. As a result, a custom built tool called Honk now handles maintenance, saving 90% of time on complex migrations.

Coding is no longer the bottle neck at Spotify; instead, figuring out what to build is. According to Gustavsson, validating ideas through prototyping has transformed how Spotify validates and prioritises ideas and prototyping that previously took days or weeks now takes minutes. He confirmed that in 6 months time, they will have a very different way of building products at the company but admitted that they’re still figuring things out as they go.

You can watch the full talk from Spotify’s Niklas Gustavsson here along with the rest of the talks from the event which are now on Claude’s YouTube channel (this talk on everything that’s new in Claude Code was also helpful).

Figma fights back with a new Agent

Reports of its death appear to be greatly exaggerated. This week, Figma released a new agent that works directly on the design canvas. The agent understands your design system, components, tokens, and team context - giving it access to information third-party tools can’t reach. It lives in the left rail alongside your layers, eliminating context switching.

The agent handles three core workflows: exploring design directions (generate multiple stylistic approaches simultaneously), automating busywork (bulk edits, component swaps, content population), and processing team feedback (summarizing comments, identifying themes, pressure-testing designs from different perspectives).

Unlike the separate MCP server that moves work between code and canvas, this agent operates within the same file where your team already collaborates. Investors seem pretty happy with Figma’s new announcements and the company’s stock is up 20% this month so far.

---

Tools you can use

Magic Path 2.0 - A canvas for product design. This second version is now a multiplayer canvas for humans and agents like Codex or Claude Code to design and build with AI.

A browser CLI - gives Claude and other agents 200+ pre-built integrations from weather.gov to LinkedIn to recreation.gov, so they can search, scrape, extract, and automate tasks across the open web without writing code.

Prelude - improve your product’s onboarding KPIs with world class human verification and authentication. Raised $20m this week.

---

📈 Product data and trends to stay informed

The latest State of AI in product design report found that more designers than ever are using AI to generate code. 50% of respondents said they are using AI for code generation - up from just 19% last year.

Here’s a snapshot of the core AI use cases for design teams from the report with a YoY comparison:

Other nuggets of data from the report worth knowing for product teams include:

• The average AI toolstack has gone from 3 tools to 7 in a year

• 78% use Claude as their primary AI tool - overtaking ChatGPT (65%) and 65% use Claude Code - a tool that didn’t exist when the 2025 survey ran

• A massive 74% of designers at 2,000+ employee companies use internally built AI tools

Google’s AI Overviews are now used by more than 2.5 billion monthly users, while AI Mode (the conversational search launched last year) now tops 1 billion monthly users. For comparison, ChatGPT has 900 million weekly active users:

OpenAI has released a B2B version of its Signals report which looks at how companies are using its set of products. It includes some interesting data points for product teams, including one nugget which explains that the gap between high AI adoption companies (frontier companies) and the rest is big - and is growing.

The report confirms that the frontier firm demands 3.5x as much intelligence per worker as the typical firm. This gap has increased from 2x in April 2025, suggesting that firms using AI most deeply are widening their lead and are better positioned to translate new AI capabilities into deeper, more complex work.

More tech firms are using their own employees as AI training data sources. Microsoft collects code written by its 100,000 developers through VSCode and Xbox game source code, then tracks which AI-generated suggestions engineers approve. Meta takes a more invasive approach - its Model Capability Initiative tracks employee mouse movements and browser activity to teach AI agents how people actually use computers. xAI offered employees $420 to donate their tax returns for Grok training and BCG is building an AI agent based upon the behavior of their sales teams.

Anthropic and OpenAI now generate 89% of top AI Startup revenue with almost $80 billion in annualized revenue - up 112% from six months ago. Salesforce alone is on track to spend $300 million on Anthropic’s models, representing almost 5% of the $6.7 billion Salesforce total cost on revenues last year.

Claude’s head of product, Cat Wu, says that in the past year, the median user of Claude Code went from using it 20 minutes a day to 20 hours a week:

Benedict Evans publishes a biannual report on the state of AI. The latest AI Eats the World report gives us an update on the current state of play.

---

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.