Descriptions:
Radek Sienkiewicz, an OpenClaw maintainer who goes by @velvetshark-com, walks through how he incrementally gave an AI agent near-complete access to his digital life — including emails, calendars, files, operating system automations, and a 3,000-page Obsidian knowledge vault he has built up over years. The key point of the talk, delivered at the AI Engineer conference, is that none of this happened in a single leap: each new capability was added only after the previous one was stable and well-understood, making rollback straightforward whenever something broke.
The resulting setup spans five functional areas: ambient operations (overnight indexing, backups, version updates with pre-flight verification scripts), attention filtering (proactive alerts based on cross-referencing incoming emails against Obsidian context), knowledge work assistance (surfacing relevant past notes when adding new bookmarks), calendar and communication management, and creative/research workflows. Sienkiewicz notes his architecture closely matches what Andrej Karpathy described in a viral tweet about personal LLM knowledge bases — something he arrived at through iteration rather than design.
The talk is less a technical teardown of OpenClaw internals and more a methodology for safely scaling agent access over time. Sienkiewicz’s incremental approach — see a need, solve it simply, add one step, observe what breaks — offers a practical counterpoint to both the fear of agents causing irreversible damage and the hype around fully autonomous AI systems.
📺 Source: AI Engineer · Published May 02, 2026
🏷️ Format: Workflow Case Study
