Summary
Nick Saraev walks through a concrete, low-code system he built around Fable — Anthropic’s computer-use model — to passively monitor his own screen and surface workflow inefficiencies. The setup is straightforward: Fable watches the screen and captures a screenshot every five seconds, logs the frames to a daily folder, passes them through a cheap summarization model, tallies repeated low-value behaviors, and then queries Fable periodically with the accumulated ledger asking how to economize the workflow.
Saraev documents the real gains he saw in the first few days: Fable identified that he was polling several websites in a loop and proposed a batched morning digest instead; it noticed a multi-step idea-capture routine and replaced it with a single Raycast hotkey backed by a Linear API passthrough; it flagged a repetitive page-opening sequence and wrote a Chrome extension to handle it automatically. On the measurement side, switching from a paid voice transcription tool (Aqua) to a Fable-recommended alternative saved $15 per month and cut per-transcription latency from 250 ms to 50 ms — a meaningful aggregate saving given hundreds of daily uses.
The video frames this as a scalable onboarding and optimization primitive for organizations, not just individuals, and speculates briefly about extending the same pattern to physical-world behavior via room cameras. The system consumes roughly 8,640 lines of text metadata per session, making it practically free to run at scale.
📺 Source: Nick Saraev · Published July 09, 2026
🏷️ Format: Workflow Case Study
