Smart devices, IoT, and home automation

Physical-world hooks people mention—air purifiers, TVs, sensors—when vendors expose documented APIs or local bridges:

• Bounded automations: one room, one device class, one outcome (comfort, not “optimize my life”).
• Human overrides always available—physical switches still exist.
• Telemetry summaries (“air quality trend”) separate from actuation (“set mode”).

Skip vague “make my home smart.” Name devices, states, and risks.

• Skills wrap vendor quirks into stable commands for the household.
• Approvals for anything that could surprise someone physically present.
• Self-hosted bridges (Home Assistant, MQTT) pair with local-first installs.

1. Inventory devices: cloud vs local control; offline behavior.
2. Fail-safe defaults: if the agent is wrong, what is safe?
3. Rate-limit actuation—avoid thrashing compressors or motors.
4. Notify humans on unusual commands—potential hijack signal.
5. Document manual override for guests who do not use chat.

• Network segmentation so IoT cannot reach payroll VLANs.
• Firmware update policy—IoT is a long tail of CVEs.

1. Read-only telemetry for a week—learn baselines.
2. Single automation with rollback (scene revert).
3. Test power-loss and Wi‑Fi outage behavior.
4. Expand slowly; prefer scenes over raw per-device spam.
5. Review logs after vacations—unexpected patterns surface then.

• “What could go wrong if this command runs twice?”
• “Prefer local control path when cloud is down.”
• “Summarize energy impact of this schedule in plain language.”

• Fire and water adjacent devices reviewed by someone who understands the hardware.
• Guest instructions posted physically.
• Kids cannot trigger unsafe modes from unlocked tablets.

• Cloud APIs that change silently—automations break on vendor whims.
• Security cameras and mics—highest sensitivity, highest regret.
• Confident wrong physics—verify HVAC advice with pros.
• Notification spam from chatty sensors.