User-built AI personal operating systems

The important shift is that AI personal operating systems are being built first by obsessive users stitching together their own work graph, not by polished all in one products. In practice, that means the product is the wiring itself, email, Slack, calendar, call transcripts, bugs, and browser actions pulled into one chat surface, with real value coming only after repeated testing, repair, and refinement across dozens of narrow automations.

• This workflow looks more like a self managed Zapier plus executive assistant than a single super model. The operator runs about twenty daily automations, uses Codex as the action layer across apps, and calls Claude for harder judgment tasks like strategy, code review, and big UX decisions.
  1 sacra
• The bottleneck is not raw model intelligence, it is integration reliability and memory. A large share of usage goes to keeping auth alive, expanding API access, checking broken connections, and re injecting goals and preferences that still do not persist well enough across sessions.
  1 sacra
• This is why the category is filling from multiple entry points. Tasklet starts from app integrations and workflows, Perplexity from model agnostic orchestration for prosumers, and Replit from coding into broader white collar tasks. Each is trying to turn one sticky wedge into the control plane for daily work.
  2 sacra 3 sacra 4 sacra

Over time, these hand built stacks should collapse into cleaner products that ship pre wired connectors, persistent memory, and confidence based guardrails by default. The winners are likely to be the products that reduce maintenance work fastest, because once users trust one agent layer to see everything and act safely, it becomes the default interface for work.