đ” Claude Code to Figma and the new era of Product Taste
Plus: Google's PM on the "Only Skills that matter in 2026", Claude Gemini 3.1, Cursor plugins
Hi product people đ,
This week, there is one word thatâs seemingly on the lips of many folks in tech: âtaste.â Weâll explore some perspectives on taste from tech leaders including OpenAIâs co-founder, Googleâs AI PM, Figmaâs CEO and Linearâs Head of Product who disagrees with the consensus.
Plus, Google releases Gemini 3.1 Pro with some new skills helpful for crafting presentations, practical ideas on how to use Claude Code to Figma, and the latest data from Ramp on who is winning the AI adoption race so far in 2026.
Enjoy the rest of your Friday and have a great weekend!
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 - How to use Claude Cowork as a competitive intelligence system
Discover how to turn Claude Cowork into a practical competitive intelligence system for your product work. This guide walks through real examples of using Cowork to scan competitor pricing pages, critique their UI, and automatically generate Excel matrices, strategy decks, and Slack-ready stakeholder updates.
New in the AI library - How to develop Product Taste
As AI makes it easier than ever to bring ideas to life, many commentators are saying that taste is the ultimate differentiator. OpenAIâs co-founder tweeted this week that âtaste is a new core skillâ:
Hereâs a selection of new resources and prompts added to the DoP AI library this week to help sharpen your sense of product taste:
Essays on Taste - a collection of perspectives on developing a sense of product taste. Includes Paul Grahamâs classic on taste from 2002 as well as new perspectives from leading CEOs at companies like Figma and Linear and more.
Mastering Product Taste mini lecture - this mini lecture turns the essays on taste into a practical set of guidelines you can use to develop your own product taste.
Tasteful design feedback prompt - use this prompt to articulate your thoughts and feelings about a design. It reflects the feeling back, offers vocabulary to recognise rather than invent, and asks simple questions to draw out the thought, ending with a single sentence they could actually say to a designer.
(Department of Product)
Tools you can use - Lemon - turns your voice instructions into finished tasks
Lemon is the first Al agent that turns your voice instructions into finished tasks. Reply to messages, ask questions, and delegate tasks without ever leaving your current tab.
Process - How design leads are using AI
Design leads spend 80% of their time communicating, aligning, and justifying. Thatâs exactly where AI helps most. (Medium)
Design inspiration - Linearâs new homepage
This past week, Linear unveiled their new homepage. The company is notorious for its pixel perfect designs and this latest iteration of their homepage doesnât disappoint. It comes with a fully interactive demo of the productâs core UI that users can interact with and more. (Linear)
Case study - How Doordash built a proprietary agent to reduce menu errors by 30%
DoorDash turned their own QA process and expert judgment into a calibrated âgrader,â then used it as a reward function to train a proprietary agent that fixes messy, realâworld menus. The result: a 30% relative reduction in critical menu errors at production scale. (DoorDash)
Strategy - Why context is now a competitive advantage
Context is your real AI moat - not models, vendors, or tools. This new piece in the Harvard Business Review argues that the way your teams actually work (the invisible judgment calls, coordination patterns, and tradeâoffs buried in emails, chats, and docs) is what will separate AI winners from everyone else. (Harvard Business Review)
Skills - Googleâs Senior AI PM on the âOnly skills that matter in 2026â
This piece argues that in 2026, raw coding ability has been commoditized by AI, and the real leverage has shifted upstream to five meta-skills: problem shaping (turning fuzzy goals into precise tasks agents can execute), rich context curation, product âtaste,â agent orchestration, and knowing when not to use agents at all. (X)
New product features and innovation this week
Anthropic and Figma have launched a new Claude Code to Figma MCP integration. Users can now send the rendered state from Claude Code straight into Figma as fully editable layers by installing the Figma MCP and typing âSend this to Figma.â Once in Figma, designers can use the canvas to explore many visual options, compare them sideâbyâside, and refine details with direct manipulation. Once youâre happy with the design changes, these can then be pushed back into Claude Code for more functional manipulation.
Figmaâs CEO says that the traditional product â design workflow is changing. Instead of a traditional âbrainstorm â design â codeâ sequence, he says modern workflows can start anywhere (terminal, prompt, UI, sketch) and go everywhere. He echoed this sentiment further on the companyâs earnings call this saying that:
Weâre seeing roles and responsibilities blur across design, engineering, and product management. Figma Make is expanding use cases, bringing in diverse personas like product managers, and encouraging a more generalist approach within teams.
Figmaâs CEO says that because of this blurring of boundaries between the traditional roles in a product team, the ultimate skill is now about deciding what to build.
Earlier this week, Linearâs CEO expressed frustration with the idea that designs should start in code, saying that often this is too often to the detriment of the overall UX, but this was before Claudeâs announcement. Linearâs product manager, though, isnât convinced that humans actually possess more taste than AI models. He says that âIf you ask any leading model about a product decision with guidance: "don't tell me to ask users; just reason through it yourself." It will give you a better answer than 90% of PMs.â
đ§ Ideas for practical ways product teams can experiment with the new Claude Code Figma integration
If youâre curious about trying out Claude Code to Figma, here are some potential ideas to experiment with:
Design analytics archaeology - render legacy features from production code back into editable Figma layers, then cross-reference with analytics to understand how old designs perform and why decisions were made
Accessibility checker - run automated accessibility analysis, generate violations as annotated Figma layers with suggested fixes, then push the corrected designs back to code
A/B test design generator - feed performance data (drop-off rates, rage clicks) into an AI model, generate multiple design variants informed by real data, land them all in Figma as editable explorations, then mix and match before shipping
Multi-model design evaluator - Send the same prompt to multiple AI coding models simultaneously, render each output into Figma as separate frames on the same canvas, and compare results side by side
Airbnb tests new AI Search functionality with ultimate vision to build an ongoing AI search assistant
Airbnb is testing out a new AI-powered search feature according to their latest earnings shareholder letter. In the letter, they say they are âapplying AI to how people discover and plan trips on Airbnbâ and taking an iterative approach. Early tests focus on letting guests describe in natural language what they want and ask questions about both the listing and the surrounding location, instead of just clicking through filters and photos. Over time, they expect this to evolve into a more comprehensive and intuitive search experience that continues through the entire trip, not just the initial booking. Search is an area where some of the worldâs top tech companies are integrating AI to transform it into a more intelligent assistant. Read the latest DoP Deep on Search here.
Pinterest issues a âcode redâ, internal struggles with voice as a primary UX
Pinterest has issued a âcode redâ to revive growth and defend against AI-driven competition, triggered after weak Q4 ad revenue and rising pressure from rivals like Meta, Google and AI chatbots.
Earlier this year in their company filings, Pinterest confirmed that it was doubling down on the development of new AI features following the lay-offs affecting around 800 staff members. But according to reports, its product teams are grappling with an internal struggle about what that should actually look like. Their CEO says that the company should build a voice-powered AI assistant to differentiate itself against others in the market who largely rely on text. The introduction of voice-based search features were apparently more valuable to advertisers in early tests, with the share of commercial searches increasing 25%. But product and design leaders have pushed back against a deeper focus on voice as a primary UX, arguing that voice controls didnât fit a primarily mobile product that could struggle with background noise.
Docs gets audio, Gemini 3.1âs micro-animations could transform presentations
Google Docs is getting audio overview summaries that summaries the contents of a document across multiple tabs which could be handy for exec or other folks who just want a quick snapshot. Plus NotebookLM is getting prompt-based revisions which allow you to make edits with simple prompts and export them into Powerpoint decks.
Google also launched its latest mode Gemini 3.1 Pro, which it says can generate website-ready, animated SVGs directly from a text prompt. And because these are built in pure code rather than pixels, they remain crisp at any scale and maintain incredibly small file sizes compared to traditional video.
These might not only be useful for product design and UI demos on landing pages but also for presentations where you want to bring complex concepts to life.
Cursor gets plugins for building payment integrations in Stripe, data analytics in Amplitude, planning in Linear
Cursor just launched its own version of plugins - a way to extend its AI coding agent with connections to external tools and services.
Plugins bundle things like MCP servers, custom prompts (âskillsâ), sub-agents, rules, and hooks into installable packages. Cursorâs versions are available through a new Cursor Marketplace, and you can also build and share your own.
Some of the plugins that might be of interest to product teams include Linear for pulling in issues, projects, and documents directly into the agentâs context; Amplitude for querying behavioral data, analyzing growth dashboards, and synthesizing customer feedback; and Stripe for building and testing payment integrations end to end. Stripeâs product manager shared one example of a Cursor plugin which lets product teams use a prompt like âUse the Stripe plugin to integrate a $20 /mo subscription serviceâ, saying itâs now âeven easierâ.
đ Product data and trends to stay informed
A new report from Figma shows that the need for proficiency with AI tools in the design process has increased in the last 6 months. 76% of tech companies in the APAC region and 79% of tech companies in the EMEA region say the need for AI tool proficiency has increased.
Many hiring managers explicitly say they now screen out candidates who do not actively use AI in their dayâtoâday work:
A French tech hiring manager says they now refuse to hire designers who donât use AI daily, because theyâll be âless efficient and [take longer] to design great experiences related to AI in our products.â
A UK hiring manager says they are âhiring less people for repetitive tasks and instead strategic thinkers, willing to try and fail and fail again,
before finding success through iterative and agile team workâ]
The report also finds that itâs becoming more challenging for junior designers to land their first role with more demand for senior designers with AI proficiency. Is hiring candidates using AI fluency criteria fair?
Goldman Sachs has been working with Anthropic for 6 months to build internal tools and AI agents. The AI agents are in at least two specific areas so far: accounting for trades/transactions, and client vetting/onboarding with other areas like compliance and accounting potentially set for automation next. Finance companies are increasingly turning to tools like Claude Code and one example cited in The Information tells the story of a stock brokerage company who ditched the SaaS products they previously used to build their own versions in-house instead, leading to a $150,000 annual cost saving.
Around 69% of firms across the US, UK, Germany and Australia report currently using at least one AI technology, with the US highest (78%) and Australia lowest (59%). New study by National Bureau of Economic Research.
Atlassian has imposed a hiring freeze after its stock was hit hard by the ongoing SaaS selloff. Itâs down a massive 74% on the year.
Shopifyâs CEO is now shipping more code than ever. His GitHub profile shows the number of commits has increased 10x from 94 in 2024 to 957 today so far. Commit numbers donât mean much but whatâs interesting is the kinds of projects Shopifyâs CEO is contributing to.
Heâs contributing to projects like Liquid Language Evolution, reworking Shopifyâs openâsource Liquid engine for better performance; Ruby developer tools that speed up workflows and improve reproducibility; and personal experiments such as config tools, experiment managers, and motorsport data and telemetry apps, possibly for sim racing. Liquid is Shopifyâs templating backbone which means every shopâs theme, every merchant customization runs through it. So in a sense, heâs using his own behavior to prototype and validate the cultural transformation heâs previously demanded of the company
ChatGPT recorded its highest ever daily active user figures on February 9th with 256 million. Sam Altman said India has 100 million weekly active ChatGPT users, making it one of OpenAIâs largest markets globally.
The freelance market is being hit badly by AI according to new analysis from Ramp. More than half of businesses that used freelancers in 2022 stopped using them entirely by 2025.
Ramp is also known for its monthly AI data which is always worth a read. In their latest analysis, they say that Anthropic is having a breakthrough month, based on Rampâs data. Overall business AI adoption on Ramp rose to 46.8% of businesses in January, a new high. Within that, Anthropic usage jumped from 16.7% to 19.5% of businesses in a single month, one of its largest gains; a year ago it was only about 4% (one in 25), and now itâs roughly one in five.
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 worldâs top tech companies.