Approved, not published yet. Visible on the preview site only.

9 smart home mistakes to avoid, and what to do instead

Smart home mistakes to avoid: cloud bulbs, blind toggles, careless radios and self-updating servers. Why each one happens and how to fix it for good.

Automation8 min read

An empty sofa in a quiet room, lit by a single floor lamp

It is 3 am and a light has switched itself on. None of the four apps on the phone explains why, and someone has taped over the wall switch.

Most of that trouble comes from a short list of mistakes. We do automation work ourselves and have seen all nine, some on our own test setup. The examples use a common stack: Home Assistant as the brain, Zigbee2MQTT and a Mosquitto broker for Zigbee devices, and a few ESP32 boards for infrared, all in Docker on one small always-on computer.

Why this matters, and when not to bother

Fixing these costs a weekend and ₹3,500 to ₹15,000 at most. In return the house carries on when the internet is down, the power blinks or nobody is watching. Local control also keeps your switch-on times off a stranger's server and keeps the house working when a maker shuts its cloud.

With two smart plugs and a voice speaker, this is overkill. If you rent and move every year, or nobody at home will put up with a weekend of things half-done, buy fewer, dumber devices instead. The advice below is for a hub, a dozen or more devices, and something people would miss if it stopped.

What you need

All prices are approximate and vary by brand and seller.

  • A small always-on computer for the hub software: approximately ₹15,000 to ₹25,000 new, or ₹8,000 to ₹15,000 used, for a mini PC.
  • A Zigbee coordinator: approximately ₹2,000 to ₹4,000 for a USB stick; network-attached models cost more, roughly double, and vary by seller.
  • A short USB extension cable: approximately ₹200 to ₹500.
  • A small UPS for the computer and router: approximately ₹3,000 to ₹8,000.
  • Devices that work locally, without the maker's cloud. Check before you buy.

The nine mistakes

1. Letting the cloud sit between you and your lights

Most cheap Wi-Fi bulbs and plugs route every tap through the maker's server, so an internet hiccup turns them into decorations.

We moved a Tuya colour bulb from the official cloud integration to the community integration Tuya Local in four steps:

  1. Install Tuya Local through HACS.
  2. Fetch the bulb's local key, through the integration's cloud setup or a Tuya IoT developer account.
  3. Give the bulb a DHCP reservation on the router.
  4. Turn off automatic firmware updates in the maker's app, because an update or a re-add changes the key.

Commands now land in a fraction of a second, and the bulb still answers with the internet cable pulled.

Cloud control sends every tap around the world and back; local control stays inside the house
Cloud control sends every tap around the world and back; local control stays inside the house

2. Assuming power and internet will stay up

Power cuts reboot everything at once. A common trap: a service that writes large files starts a few seconds before its storage drive has mounted. It writes gigabytes to the system disk, hidden under the empty mount folder, and nothing looks wrong until that disk fills.

Set every container to restart: unless-stopped, and make Docker wait for the storage it needs with a systemd drop-in (use your own data path):

# /etc/systemd/system/docker.service.d/10-wait-for-storage.conf
[Unit]
RequiresMountsFor=/mnt/data

Run sudo systemctl daemon-reload, reboot once, and check with df -h that the files land on the data drive, not the system disk. Give DHCP reservations to anything other devices depend on, and put the server, router and coordinator on a UPS if power cuts are common.

3. Automating everything that moves

Motion sensors are cheap, so lights turn off while you are reading and presence sensors decide the cat is a person.

Our rule is intentional over automatic. An Aqara cube on our test setup controls one lamp: face up picks brightness, hue, saturation or colour temperature; twisting changes that value; a shake toggles the lamp. It pairs through Zigbee2MQTT and publishes an action (side_up, rotate_left, rotate_right, shake) that a Home Assistant automation listens for. The cube costs about ₹2,000 to ₹3,000.

Automate it when a wrong trigger costs nothing (a porch light on too early). Leave it manual when a wrong trigger interrupts a person (a room light off while they read). An all-off when the house is empty needs a reliable presence source; skip it if you only have one motion sensor.

4. Repeating commands that toggle

Many infrared devices have one power button: press once for on, again for off. Infrared is one way, so the device never confirms, and an automation that "retries to be safe" flips it the wrong way.

Many devices also ignore commands sent too soon after power on. What works, tested on an ESP32-driven speaker:

  • Track the state yourself with a helper, and only send power when it needs to change. The helper is a Toggle created under Settings, Devices and services, Helpers.
  • Wait after power on. Ours needs a full three seconds before it will accept the next command; anything shorter is ignored. To find your own delay, start at one second and add half a second until the next command is always honoured.
  • Repeat only safe commands. Volume up can go twice. Power never should.
  • Give people a resync button that corrects the tracked state without sending anything. In Home Assistant that is a tile for the helper whose long-press action toggles the input_boolean only, nothing else. Ours shows red for on, grey for off, and a long press fixes the colour without sending IR.
# example names, use your own helper and button entities
sequence:
  - condition: state
    entity_id: input_boolean.example_speaker_power
    state: "off"
  - action: button.press
    target:
      entity_id: button.example_speaker_power
  - action: input_boolean.turn_on
    target:
      entity_id: input_boolean.example_speaker_power
  - delay: "00:00:03"   # let it wake up before the next command

5. Treating the Zigbee coordinator as an afterthought

Every Zigbee device pairs to the coordinator, so it deserves better than whatever port is free.

  • One coordinator, one program. Zigbee2MQTT and Home Assistant's built-in ZHA cannot share a stick.
  • Keep it away from USB 3 ports and the server case. USB 3 interferes with 2.4 GHz radio, and a short extension cable fixes it cheaply.
  • Pick channels apart before pairing many devices, because changing later can mean re-pairing. Wi-Fi channel 1 overlaps Zigbee 11 to 14, channel 6 overlaps 16 to 19, channel 11 overlaps 21 to 24; Zigbee 15, 20 and 25 sit in the gaps. A free Wi-Fi analyser app shows your neighbours' channels. In Zigbee2MQTT the setting is advanced: channel in configuration.yaml.
  • Back up the coordinator state. The backup file is coordinator_backup.json in its data folder. Without it, replacing a dead stick can mean re-pairing everything.
A Zigbee coordinator on a short USB extension, kept clear of the computer
A Zigbee coordinator on a short USB extension, kept clear of the computer

6. Letting everything update itself

A container updater such as Watchtower can update itself in the same run as several others and leave the containers it stopped removed. If Home Assistant is one of them, restart: unless-stopped cannot bring back a container that no longer exists.

  • Exclude the updater from updating itself:
watchtower:
  labels:
    - com.centurylinklabs.watchtower.enable=false
  • Pin anything with a database. Write image: nextcloud:<exact-version>, not latest, and upgrade one major version at a time after a backup.
  • Some stacks must never be auto-updated at all. Mailcow, for example, is updated only through its own update script, because its parts depend on each other.

7. Leaving default passwords in place

Cameras, recorders, plugs and hubs often ship with a default login or none. The hub works on day one without any of this, so nobody goes back. In Mosquitto set allow_anonymous false and create one user per program with mosquitto_passwd, then restart it; on cameras change the admin password and switch off cloud or P2P options you do not use; for remote access use WireGuard or Tailscale on the hub instead of forwarding ports.

8. Not noticing when things quietly break

A smart home fails silently: a backup job that copies nothing still exits cleanly.

We run Uptime Kuma, which checks every service and pings our phones through ntfy. In Uptime Kuma set the heartbeat interval to 60 s and retries to 3, so a two to three minute blip stays silent. For scheduled jobs create a Push monitor, curl its URL as the last line of the job, and set the expected interval slightly longer than the period (25 hours for a nightly job). Our ESP32 boards use the same idea: a phone push after 5 minutes unreachable, and again when back.

Monitors grouped by what people notice, so a failure shows its impact at a glance
Monitors grouped by what people notice, so a failure shows its impact at a glance

9. Changing things without a way back

The only copy of a config is the live one, and a bad change overwrites it in place. Use Settings, System, Backups in Home Assistant, or tar the config folder, named with date and reason. To test a backup, extract it into a spare folder and open configuration.yaml. Then change one thing at a time.

How to check your own setup

  1. Pull the plug. Switch off the server and router together for a minute. Everything should return untouched; note whatever did not.
  2. Cut the internet only. Unplug the line and use your lights. Anything that stops is cloud-dependent.
  3. Check the coordinator. Its port, its position, and its backup file in last night's backup.
  4. List your toggle devices. For each, point at a tracked state and a wake-up delay.
  5. Check update settings. The updater should be excluded, and every database-backed image should name an exact version.
  6. Break a monitor on purpose. Stop one non-critical service and wait for the alert, then the recovery message.
  7. Restore one backup to a spare folder and open it.

What goes wrong

  • A local device goes unavailable. Someone re-paired it in the maker's app, so it has a new local key. Fetch it and update the integration.
  • Everything MQTT goes quiet after a config change. Mosquitto in a container may not pick up new users on a reload signal. Run docker restart mosquitto after changing its users or config.
  • Containers missing after an update night. Run docker compose up -d from the folder that holds the compose file.
  • Hard-wired switches or relays misbehave. Any change on the mains side is a job for a licensed electrician.

Cost and time

Most fixes cost only time: an afternoon for the plug-pull test, another for monitoring and backups. Hardware (a UPS, an extension cable, a better coordinator) comes to roughly ₹3,500 to ₹15,000, and nothing at all if you already own a UPS. Replace cloud-only devices as they fail.

In short

A smart home should feel like a well kept house, not a hobby.

If yours has started to feel like a hobby, see how we approach home automation or get in touch.

Written by Siddharth Agrawal, Founder and Director, TDH Systems.

Tell us what you have in mind. We will come back with a plan and a quote.

Start a project

Prefer email? Write to info@thydarkhour.com.