The End of Apps — Kitze, Sizzy.co

Key Concepts

• Life OS: A centralized, integrated system for managing tasks, habits, events, and personal data.
• Agentic Workflow: Moving from manual prompting to autonomous agents that handle tasks in the background.
• Local-First Computing: Prioritizing self-hosted data and local models to ensure privacy and reduce reliance on cloud-based API credits.
• Multi-Agent Orchestration: Using specialized agents for different life domains (work, health, personal) rather than a single, overloaded assistant.
• Context Injection: A methodology for maintaining agent memory by nesting topics and injecting parent-level descriptions into prompts.
• UI-on-the-Fly: The future vision where software interfaces are generated dynamically by AI based on the task, rather than relying on static, pre-built applications.

1. The Evolution of Personal Productivity

The speaker traces his journey from a 10-year-old using paper checklists to building complex, integrated systems. His progression highlights a recurring struggle with "friction"—the effort required to input data into various apps.

• Early Iterations: Used tools like Tasker (Android) for contextual automation (e.g., triggering reminders based on Wi-Fi connection or location).
• The "Benji" Project: An attempt to build an "all-in-one" life OS. The project stalled due to "feature creep" and the speaker's aversion to marketing, illustrating the common trap of building tools instead of using them.
• The Friction Problem: The speaker notes that he oscillates between periods of intense logging and total abandonment of his systems due to the manual effort required.

2. The AI Agent Shift

The speaker discusses the transition from traditional SaaS apps to LLM-powered agents.

• The "Chat GPT Moment": Initially, the speaker believed LLMs would replace all standalone apps. However, he found that early models required "bullying" (complex prompting) to return structured data like JSON.
• OpenClaw & The "Lobster" Cult: The speaker became deeply involved in the OpenClaw community, which uses LLMs (like Claude) via messaging platforms like Telegram or Discord to manage life tasks.
• The Limitations of Current Platforms: He argues that Discord and Telegram are "cope" solutions—they are not designed to be operating systems. They suffer from unreliability, poor memory management, and "oatmeal-like" personality traits in newer, more restricted models.

3. Methodology: The "Wolffer" Experiment

Frustrated by the limitations of existing agent frameworks, the speaker began developing Wolffer, a personal experiment in agent orchestration.

• Design Philosophy: Unlike general-purpose agents, Wolffer focuses on predictable UI and nested context.
• Context Management: Instead of relying on a "magical" memory system, Wolffer uses a hierarchical structure. When a user interacts with a sub-topic (e.g., "Benji Customer Support"), the system automatically injects the descriptions of all parent topics into the prompt, ensuring the agent has perfect context without needing to search a database.
• Tooling: The system provides a visual interface for managing agents, viewing tool calls, and handling cron jobs, moving away from the "black box" nature of slash-command-based bots.

4. Key Arguments & Perspectives

• The Inversion of Prompting: The speaker predicts that the future of productivity will be an "inverse" model. Instead of humans prompting AI, the AI will ingest life data (emails, notifications, tasks) and "prompt" the human with decisions or next steps.
• The Death of Consumer Apps: He argues that for the average user ("normies"), the need for specific apps will vanish. Instead, they will interact with a unified OS that generates the necessary UI on the fly.
• The "Tinkerer" vs. "Mass Market" Divide: He distinguishes between tools for tinkerers (self-hosted, complex, high-control) and tools for the masses (Apple/Google-integrated, local, "it just works"). He believes Apple is well-positioned to win the mass market by giving local agents access to system-level APIs.

5. Notable Quotes

• "I realized that I never wanted a to-do app. I wanted like sort of like a life OS."
• "I don't know how the internals of my setup work. I just ask... to fix it, to change it... but I have no freaking idea."
• "The fully productive people will be the one who delegate 99% of the stuff to the AI and then the AI prompts you."

6. Synthesis and Conclusion

The speaker concludes that while the current state of AI agents is chaotic and often unreliable, the trajectory is clear: we are moving toward a world where software is invisible and task-oriented. His personal journey—from building complex, self-hosted systems to realizing the need for predictable, context-aware orchestration—serves as a blueprint for the next generation of personal computing. The ultimate takeaway is that context is more important than memory, and the future of productivity lies in systems that proactively manage our lives rather than waiting for manual input.