The End of Apps

Kitze traces his search for a workable productivity system from childhood lists to Android text files and voice capture through Google Home. He built Benji to bring many productivity features together, but says logging everything became tedious and the app’s input forms remained clunky.

After ChatGPT’s plugin announcement, Kitze imagined replacing app forms with conversational interfaces. Later, experiments with Claude Code and OpenClaw agents included email triage, smart home control, booking a snow shoveler, and collecting personal data in Markdown. The experiments felt powerful, but maintaining OpenClaw became a full-time job and many people in his community found agents unreliable and difficult to set up.

Kitze argues that one-to-one agent chats do not fit people juggling many tasks, and that a general-purpose agent lacks clear specialization. He favors named agents with their own models, instructions, tools, permissions, and skills. He also found chat platforms confusing for managing work, while the choice between self-hosted and cloud agents leaves tradeoffs around setup and data privacy.

Kitze describes Mind, a browser organized around to-dos, and Defold, an interface combining to-dos, notes, and chats. He sees agents creating interactive widgets and temporary apps for tasks such as choosing flights or packing for a trip. He says Exeutor connects MCPs across agent setups, while Skillbox makes skills portable.

Kitze predicts computers will handle most task work in the background and prompt people when they need to make a decision or unblock a task. He expects many consumer apps to give way to conversational computer interfaces, while specialist creative software remains. He speculates Apple could benefit if local models on its hardware become capable enough for everyday productivity.