Week of December 20, 2025

This Week's Top Videos

What does AI mean for education?

By Anthropic

Anthropic found 47% of student interactions with Claude are direct, transactional homework help with little learning engagement—despite AI's potential as Socratic tutors. Education teams are now racing to design AI tools that augment rather than replace human thinking. This matters now because AI education tools represent a massive market opportunity if builders can solve the engagement problem.

• 47% of student interactions on Claude were direct, transactional with little engagement, revealing users default to homework completion rather than learning
• AI tools should augment teacher connection time rather than replace it—focus on reducing administrative burden so educators can spend more time understanding students
• LLMs weren't built for education but show emergent tutoring capabilities, creating both massive opportunities and risks for student learning outcomes
• Anthropic created a dedicated 'Ministry of Education' team to study both benefits (democratized tutoring, reduced burnout) and risks (cheating, dependency) of AI in learning
• The key design challenge is creating AI tools that enhance critical thinking rather than replacing it—requiring intentional product decisions around user interaction patterns
• AI in education is forcing institutions to address long-standing academic problems they've been avoiding, creating urgency around fundamental learning assessment reforms

The best company building insights of 2025

By First Round Capital

First Round reveals why the strongest founders often get stuck with weak product-market fit—because they can sell anything. One founder spent 4.5 years rebuilding from scratch 8 times, while another fired Netflix as their #1 customer to serve 200 smaller ones. These counterintuitive moves separate billion-dollar companies from lifestyle businesses.

• Strong founders can sell weak products, creating false product-market fit signals that trap them in mediocrity instead of finding true breakthrough opportunities
• One founder rebuilt their entire codebase 8 times over 4.5 years before launching publicly, ignoring advice to ship early because they knew their space required perfection
• AppDynamics fired Netflix as their #1 customer after 3 years because serving one huge client prevented them from building for their other 200 customers
• A founder tested pricing by setting the highest price with minimal features to see signup rates—this became their best pricing validation method ever
• Mercury saw someone transfer $1 million without any sales contact 4 days after launch, proving enterprise self-service was possible without traditional sales teams
• Living in Shenzhen for 1.5 years instead of the planned month taught hardware founders that proximity to manufacturing is non-negotiable for speed

Claude's Agent Mode was LEAKED (First Look)

By Greg Isenberg

Claude's leaked Agent Mode replaces text boxes with structured workflows—research, analyze, write, build, do more—each with customizable outputs and progress tracking. This interface design could eliminate the "blank page problem" that makes current AI tools daunting for most users. For builders, this signals the shift from chat-based AI to task-delegation interfaces that feel more like managing teammates than prompting bots.

• Claude's Agent Mode introduces 5 structured buckets (research, analyze, write, build, do more) with customizable outputs, replacing the intimidating blank text box interface that makes current AI tools daunting for users.
• The interface includes a progress tracker and context manager on the right side, allowing users to monitor delegated tasks like checking in on teammates in Slack rather than having open-ended conversations.
• Hyrox fitness competition shows 5,525% search growth over 5 years with low competition and cheap CPC, indicating massive business opportunity for mobile apps, tracking tools, or community platforms.
• The research function lets users choose search sources (internet, MCPs, connectors) and customize depth (quick vs thorough), making AI research more controlled and targeted than current chat-based approaches.
• Agent Mode represents the future of work where users delegate distinct tasks to AI agents rather than having conversational exchanges, positioning AI as autonomous tools for structured workflows.
• When spotting trends with massive growth and low competition like Hyrox, the opportunity lies in creating mobile apps for tracking, recommendations, or daily check-ins before the market gets saturated.

This was the biggest debate in design this year

By Designer Tom

A Twitter debate between Linear's co-founder and Cursor's design lead revealed the hidden 2x2 matrix reshaping how designers work: floor vs. ceiling breaking (traditional) crossed with unconstrained vs. material exploration (the new axis). The key insight: it's not whether to design in code or Figma—it's knowing when to operate in each quadrant. This matters now because design tools are blurring together and the old categories no longer apply.

• Design exploration has two axes: the traditional floor/ceiling breaking, plus a new axis of unconstrained vs. material exploration that determines when to introduce constraints in your process.
• Early design benefits from 'tools that let you change your mind without paying a tax for it'—speed and looseness before asking if something is buildable.
• Material exploration means designing in actual code constraints where 'the truth only reveals itself once you start to build' and reality can push back against your ideas.
• Use 'discovery code' not production code for exploration—messy spaghetti code that you throw away after validating if an interaction has potential.
• Cross-medium prototyping (like 3D printing interfaces or physical prototypes) helps explore form before deciding how something gets built or if it even can be built.
• The most effective designers move cross-quadrant frequently, letting unconstrained exploration feed into material exploration in an ongoing dialogue.

Figma Design and Draw: 2025 features in action

By Figma

Figma's 2025 AI image tools let designers isolate objects, remove backgrounds, and expand images directly in the design flow—no Photoshop context switching. Combined with new glass effects and responsive grids, designers can now create production-ready prototypes without ever leaving Figma. This matters because it eliminates the designer-developer handoff friction that kills product velocity.

• AI-powered 'isolate' tool lets you lasso select any part of an image and separate it into its own layer, enabling complex text overlays and layered compositions without external tools.
• Remove background feature automatically isolates subjects from images with one click, perfect for creating layered hero sections where text sits behind foreground elements.
• AI expand feature uses machine learning to intelligently extend images to fill different aspect ratios while maintaining visual consistency across responsive grids.
• Glass effect with customizable frosting, dispersion, and lighting creates modern UI elements that maintain readability over complex backgrounds in scrolling interfaces.
• Edit with prompt feature transforms existing images into different styles by referencing other imagery, maintaining brand consistency while generating new assets.
• Responsive grid system with hug properties automatically adjusts layout as content is added, enabling dynamic layouts that work across devices without manual resizing.

How To Transfer Files Between Any Device for FREE

By Kevin Stratvert

LocalSend enables instant file transfers between any devices (Windows, Mac, Linux, iPhone, Android) without cloud uploads or accounts—files go directly device-to-device over local WiFi. This open-source tool eliminates the productivity tax of emailing files to yourself or waiting for cloud sync, giving builders faster workflows for moving assets between development machines.

• LocalSend works across all major platforms (Windows, Mac, Linux, iPhone, Android) with automatic device discovery on the same WiFi network
• Files transfer directly device-to-device over local WiFi without any cloud storage, accounts, or data uploading to external servers
• Local transfers are significantly faster than email or cloud storage methods since files don't need to upload and download again
• The tool can transfer multiple content types including photos, videos, text, clipboard content, individual files, and entire folders
• Browser-based version available at LocalSend.org for quick transfers without installing apps on locked-down or shared computers
• Open source and completely free with no ads, making it a reliable tool for professional workflows