AI Builders
Digest

The Takeaway: Marketing has shifted from optimizing for human search to making your brand legible to superintelligent AI agents that now do the browsing and deciding.

James Cadwallader, co-founder and CEO of ProFound, the AI-era marketing platform serving nearly 12% of the Fortune 500, explains that we're in the midst of the biggest platform shift in marketing history—from blue link search to probabilistic AI responses. The key difference is that agents, not humans, are now the ones crawling the web, using far more of the long tail (like 65 pages for a simple product query) and remixing information.

The biggest misconception is treating agent optimization as just SEO 2.0. In reality, content is now designed to be discovered and consumed by agents, not humans. 'You are building content that may quite literally never be consumed by a human.' To stand out, marketers need original insights because 'humans are this kind of fleshy API between reality and the Internet.' Cadwallader predicts a possible 'dead Internet' in three years if incentives for human content creation collapse, pushing innovation to social media.

For the future, he sees hyper-personalized generative ads via system prompts and urges marketers to use agents like ProFound's to not just monitor visibility across models but to build entirely new marketing capabilities that weren't possible before.

Box CEO Aaron Levie explains that setting up AI agents at company scale takes real work, meaning the role of the forward deployed engineer isn’t going away soon. When vendors sell agents into organizations, they’re selling workflow transformation rather than just a software tool, requiring deep domain understanding, system wiring, context setup, and change management. This creates opportunities for professional services firms as new practice areas emerge in system integrators and consulting firms to handle agent adoption.

FPV Ventures partner Nikunj Kothari pushes back on 'vibe code this in a weekend' claims by recalling the complexity of systems like Slack's notification system. While simple workflows may be vibe coded away, quality, depth, and great systems still require time, persistence, and effort. You can’t vibe code lessons learned from building robust products. Now and forever.

South Park Commons General Partner and ex-Dropbox CTO Aditya Agarwal observes that every day he sees someone making LLMs do something weird and interesting. He reminds himself that we are still in the super early innings of figuring out what we can build with this magical technology.

Replit CEO Amjad Masad announced that Indian Replit Builders can now build and pay with Razorpay. He also shared that building one-off conversion apps has become a skill for him.

Every CEO Dan Shipper introduces Sparkle v4, an agent that examines your filesystem to create a personalized organization structure, performs deep cleans of junk, and runs quietly in the background on a schedule to keep your Mac organized. He credits his team @poojary_yash for the incredible work, calling it one of the best apps in the Every subscription.

Y Combinator President and CEO Garry Tan released GBrain as free, open source under MIT License. Built in 12 days using OpenClaw and his GStack toolkit, his personal version now holds over 17,000 pages and truly knows him.

Claude Blog: Redesigning Claude Code on desktop for parallel agents

Anthropic has redesigned the Claude Code desktop app to support the new reality of agentic coding: developers running multiple tasks across different repos simultaneously, with the user as orchestrator. The update includes a new sidebar to manage all active and recent sessions, drag-and-drop layout for arranging workspace with terminal, file editor, diff viewer, and previews.

New features allow reviewing and shipping code without leaving the app: integrated terminal for tests/builds, in-app file editor for spot edits, faster diff viewer, and expanded previews for HTML/PDF. It maintains parity with CLI plugins, adds SSH support for Mac, and offers view modes (Verbose, Normal, Summary) plus new keyboard shortcuts.

'The new app is built for how agentic coding actually feels now: many things in flight, and you in the orchestrator seat.' Available now for Pro, Max, Team, and Enterprise users via download or update. Direct link: https://claude.com/blog/claude-code-desktop-redesign

Anthropic's Claude Code and Cowork engineer Cat Wu shares that she's a daily user of Claude Code on desktop, the best way to kick off and manage multiple sessions across local and cloud with git status, pinned sessions, and drag-and-drop layout to view multiple at once. She also announces Routines, allowing users to automate Claude Code on a schedule, from GitHub events, or other APIs.

Anthropic Research's Alex Albert highlights the new version of Claude Code in the desktop app with tons of new features and performance improvements. He notes that between Cowork and Code, he no longer needs to open other apps or even the terminal for most of his work.

Builder Zara Zhang details her process of creating Tab Out, a Chrome extension that turns the new tab page into an entry point for managing open tabs grouped into 'missions'. She learned to cut 80% of features since Claude tends to overkill, resulting in a pure local Chrome extension with no server.

Vercel CEO Guillermo Rauch advises that if you're building an agent coding platform, pair app generations with a highly elastic, Postgres-compatible database service like DSQL.

Roblox product Peter Yang enjoys using the new Claude Code app and questions how to sync sessions across desktop and mobile. He shares Dylan 's insight that AI output should be treated as clay to mold with your taste, as 'you're the judge of the system' when exploring with agents—AI gets you to average fast, but your taste pushes past it.

Google VP Josh Woodward, overseeing Gemini App and AI Studio, announced new NEET practice tests in Gemini for India's medical exam. The team is adding more practice tests for different subjects and countries.

Linear head of product Nan Yu reminds critics that Google and Meta are unfathomably large companies. Their misses seem enormous because their basic unit of execution is huge, advising to count wins not losses in most businesses.

FirstMark Capital VC Matt Turck notes that while AI gives every VC the same investment advice, their independent thinking means there's no risk of herd mentality.

OpenClaw contributor Peter Steinberger expresses excitement over a new release he wasn't involved in, crediting maintainer Vincent Koc and the team. He's preparing for a TED Talk in Vancouver and will soon return to OpenClaw work.