Writing on agents, automation, and the trust work behind useful tools.
All posts
- 01 On-device AI Models with a Harness
Chrome's on-device Prompt API can power serve as a good LLM for small tasks. How does it do with tools?
- 02 LimitBar: A macOS Menu Bar for AI Usage Limits
LimitBar turns Claude, Codex, Gemini, and OpenRouter limits into glanceable battery icons, reset alerts, and live token activity.
- 03 FamilyClaw: An AI Agent for the Boring Parts of Family Life
A Budapest weekend test showed where a family AI agent earns its keep: research, trade-offs, follow-ups, and calendar coordination.
- 04 Personal AI Assistants Are Worth the Setup
The hard part is not the model. It is identity, memory, permissions, tools, and a workspace where the assistant's work stays visible.
- 05 Google AI Search Is Replacing the Map
AI search is changing the contract of discovery from finding sources to receiving answers, and that changes what content is worth building.
- 06 AI Graphics Need Systems, Not Vibes
Image models are useful raw-material generators, but consistent graphics come from arguments, layout systems, deterministic rendering, and taste.
- 07 The Approval Layer Is the Product
AI automation only becomes useful when human review, permissions, correction, and auditability are designed into the workflow from the beginning.
- 08 Marketing Ops Is Becoming Systems Engineering
Marketing operations is shifting from campaign support into system design across data, automation, AI judgment support, and operating loops.