# DHH says Mac, Windows are unfit for agents _DHH told a16z agents flipped Linux flaws into strengths, letting a tiny team rebuild Omarchy live and work 16-hour days._ **Published:** 2026-11-20 **Source:** https://www.startuphub.ai/ai-news/technology/2026/dhh-says-mac-windows-are-unfit-for-agents --- [David Heinemeier Hansson](/startups/david-heinemeier-hansson) told [a16z](https://www.youtube.com/watch?v=TtOoLanFzGw) he is working 16-hour days because agents let a tiny team rebuild a desktop OS live. Mouse-driven systems lock that power out, Hansson said, because each change still needs clicks. His fix is [Omarchy](/startups/omarchy), a ready-to-use desktop he calls the first native OS for agents. Agents, software that operates a computer to finish multi-step tasks, love what humans hated about Linux, Hansson said. Terminal commands, plain config files and verbose errors give them handles to grip. He pointed to hotkeys on Mac, where remapping workspaces means opening Settings and clicking field by field. On Omarchy a user says move the clock or build a new start menu, and the agent edits the system in seconds or minutes, Hansson said. Hansson is the creator of Ruby on Rails and co-owner of Basecamp. He said Apple now warns users against giving tools like [Claude](https://www.startuphub.ai/ai-news/artificial-intelligence/2026/anthropic-unlocks-ai-economic-data) file access, which starves agents. Omarchy ships with one installer and strong defaults on an Arch Linux base, with agents to debug issues, according to its site and guides. A community member rebuilt Hansson’s last Mac-only holdout, Adobe Lightroom, in about two weeks, Hansson said, and made it better. Limits remain human, not token budgets. Hansson said he has not hand-written a line of Omarchy code in five or six months, but has directed hundreds of thousands of lines through models from Opus 4.8 to [Opus 5](https://www.startuphub.ai/ai-news/artificial-intelligence/2026/anthropic-launches-claude-opus-5).5. Most weeks he runs on one 200-dollar, 20x subscription, Hansson said, and he rarely maxes three such plans. The grind is taste, he said, because he lives for the last 2 percent of polish and agents still need that eye before work ships. Speed is the tell to watch. Hansson cited a demo running at 1,500 tokens a second, chunks of text models read and write, that turned a prompt for Tetris into a playable game in four seconds, then mutated it live. Current tools feel frozen at about 50 tokens a second, Hansson said, but he asks what happens at 100,000 or a million. If that arrives, he said, desktop falls first, then phones, and true superintelligence could design the machine itself next year. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory. © StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training on this content requires a license. See https://www.startuphub.ai/terms.