AI Builders
Digest

The Takeaway: True creativity thrives when you set firm boundaries against constant connectivity and use AI as a powerful research and infrastructure partner rather than a writing collaborator. Writer and builder Greg shares how he rebuilt expensive SaaS tools like Quicken, Campaign Monitor, and even a custom Twitter-like space called The Good Place for his members using Claude. Everything in The Good Place disappears after a week, posts are limited, and it has no algorithm—creating a refreshing, ephemeral digital gathering spot that members now join just to access. Greg maintains deep focus by using a dedicated offline writing laptop, avoiding his phone until after lunch, and treating AI as a research assistant for fact-checking and cultural sensitivity rather than touching his prose. He embraces the "dorkiness" of building with LLMs, noting most usage is mundane while builders like him create n-of-one software that feeds back into his writing practice. A memorable insight: "If I don't create these barriers, I will lose connection with this other part of me. That I think is the most valuable part of me."

Grok 4.5 is now available to all Vercel customers and integrated into Cursor, with strong performance in complex knowledge work. Replit CEO Amjad Masad questions when we'll stop comparing autonomous agents to hand-written code, noting compilers aren't compared to hand-writing assembly. Box CEO Aaron Levie highlights how latest models excel at sophisticated domains like legal and healthcare, praising Grok 4.5 on cost and performance. Cursor's design lead Ryo Lu invites users to try it in the editor.

Anthropic addressed recent quality reports by fixing reasoning effort defaults, session memory bugs, and verbosity prompts in Claude Code. New /checkup command helps clean skills, dedup docs, and optimize setups. Managed Agents now support self-hosted sandboxes and MCP tunnels for enterprise security, decoupling brain from hands for better scalability and control. Claude Tag enables multi-player agent collaboration.

Boris Cherny from Anthropic announced /checkup in Claude Code which cleans unused skills, dedups CLAUDE.md files, breaks up monolithic docs, turns off slow hooks, updates versions, and pre-approves commands. It confirms changes before applying. Thibault Sottiaux notes OpenAI shipping mode with late-night orders. Cat Wu teases multi-player Claude Tag live walkthrough.

Greg rebuilt tax software, newsletter tools, and a custom social platform using AI coding. Trq212 notes rewrites are now good, cheap and fast. Swyx praises hand-drawn slides for humanity. Builders are creating ephemeral, low-engagement social spaces as alternatives to algorithmic feeds.

Replit CEO Amjad Masad asks if they should add CAD 3D modeling capabilities. This aligns with broader trends of making sophisticated tools accessible through AI-augmented platforms.

Greg's membership program now includes custom software like The Good Place social space and board meeting archives, driving new signups purely for tool access. This demonstrates how writers and creators can use AI coding to build differentiated value and reduce reliance on expensive incumbents like Campaign Monitor (dropping from $67k/year to ~$150).

Guillermo Rauch announced Grok 4.5 availability for all Vercel customers, emphasizing AI making all software native with uncompromising performance.

Zara Zhang observes that heavy Codex usage leads to fewer human meetings and weakened team culture, calling for human-human-agent collaboration. Nikunj notes pendulum swings between models and polished content correlating with slop. Matt Turck comments on World Cup of AI content evolving from slop to impressive.

Sam Altman reacts to various videos and shares humorous observations about models and dating.

Linear's Nan Yu shares insights on storytelling for multiple audiences and power laws from Glengarry Glen Ross.

Anthropic's Managed Agents now allow running sandboxes in customer-controlled environments via providers like Cloudflare, Daytona, Modal, and Vercel, plus MCP tunnels for private services. This decouples agent brains from execution hands for better scalability and control.