Home Assistant After 18 Months: What I Stopped Automating and Why

Last November I sat down to do a proper audit of my Home Assistant instance before migrating to a new mini PC host. I expected to find maybe 40 automations. I found 67. I could only explain the purpose of about half of them without opening the YAML. The rest were things I’d built at 11pm on a Saturday because I could, not because anyone actually needed them. That audit turned into a two-week cleanup, and this post is what came out of it.

The Setup, For Context

I’m running Home Assistant OS on an N100-based mini PC here in Calgary. Zigbee network through a Sonoff Zigbee 3.0 USB dongle via Zigbee2MQTT, Z-Wave through a Zooz 800 series stick, and a mix of Shelly devices on WiFi for higher-load switching. Roughly 140 entities across two floors plus the garage and a detached workshop. I work from home three to four days a week, so home occupancy is genuinely complex â it’s not a simple “everyone leaves at 8am” household.

I’ve been using Home Assistant since before the Lovelace UI was default, so I’m not a newcomer. That context matters because a lot of the automation sprawl I built up wasn’t inexperience â it was overconfidence.

The 12 Automations I Deleted (And Why They Failed)

Automations that made logical sense but annoyed real people

Auto-dim living room lights after 9pm based on TV state: Technically flawless. My wife hated it every single time. Deleted after six weeks.
Turn off workshop lights if no motion for 20 minutes: Turned off the lights while I was under a vehicle not moving. Twice. Deleted immediately.
Lock the front door automatically at 10pm: Locked my father-in-law out at 10:02pm the one time he visited. Deleted that night.
Notify me when the dishwasher finishes: I built this using a power monitoring plug and a threshold trigger. Took me four hours. My wife checks the dishwasher herself. Nobody used the notification ever.
Good morning routine that started coffee maker, opened blinds, and played a weather briefing: Worked perfectly for about nine days. Then one of us had a day off and the briefing played at 6:30am anyway. Deleted.

Automations that were pure theatre

Presence-based welcome lighting scene when arriving home: The Zigbee bulbs in the entryway took 800ms to respond. By the time the scene ran, I’d already walked past the entryway. The lights changed for nobody.
Christmas tree auto-on at sunset: Seasonal, so I rebuilt it every December. The outlet died. Never replaced it. Four years running.
Bathroom fan timer automation triggered by humidity sensor: The SHT30 sensor I used drifts badly in a humid bathroom environment. It would trigger the fan during showers that had already ended and not trigger during showers that needed it. Replaced with a plain $12 in-wall timer switch. Problem solved without Home Assistant involved at all.
Security alert if a door is open more than 5 minutes: Calgary in July â every door is open for 20 minutes constantly. Suppression conditions got so complex the automation stopped being auditable. Deleted.
Garage door reminder if left open after sunset: My phone notification fatigue was already bad. This notification got ignored along with everything else. Deleted as part of a broader notification audit.
Washer done notification via power monitoring: Same story as the dishwasher. Built it. Demonstrated it. Never used again.
Auto-adjust thermostat based on outdoor temperature via Environment Canada API: Environment Canada’s XML feeds are unreliable. The automation would pull stale data and do nothing, or occasionally do something wrong. The Ecobee handles this better natively. Deleted.

The 5 That Actually Work and Why

They solve a real, recurring friction point

The automations I kept have one thing in common: someone in the house noticed when they stopped working. That’s the only test that matters.

Workshop heat on a schedule with manual override: The workshop is in a detached garage and heating it takes 45 minutes to get comfortable. An automation pre-heats it at 7am on weekdays when I’m likely to use it, with a simple dashboard toggle to disable it the night before if I know I won’t be out there. This one I’d rebuild from scratch tomorrow if I lost it.
All lights off when everybody leaves: Uses a combination of phone GPS presence and a manual “we’re actually gone” input boolean to avoid false triggers. Simple, reliable, and has saved us from leaving lights on overnight a dozen times. The input boolean requirement means it only fires when I’ve explicitly confirmed we’re leaving â no more “the automation thinks we left but someone is still home” problems.
Basement dehumidifier automation in summer: Runs the dehumidifier only when the basement humidity is above 58% AND the outdoor dewpoint from my weather station is below the indoor humidity. This prevents the dehumidifier from running when it physically can’t help. Real logic, not just a threshold trigger. It’s cut our summer dehumidifier runtime by roughly 30% compared to just running it on a schedule.
Porch light on at sunset, off at 11pm, brief on again at 6am: Boring. Not impressive. Works perfectly every single day. Sunset time adjusts automatically. Nobody thinks about it. That’s the point.
Notify if the side gate has been open more than 10 minutes between 8am and 8pm when we’re home: The time window and presence condition eliminate almost all false positives. We have dogs. This has caught the gate blowing open three times and a service worker leaving it open twice. Genuinely useful.

What Automation Theatre Actually Costs You

I want to be direct about something that gets glossed over in most Home Assistant content: complexity in your automation stack has a real maintenance cost, and that cost compounds.

Every automation you build is something you have to debug when Home Assistant updates, when an integration breaks, when a device falls off the network, or when your household routine changes. Sixty-seven automations isn’t impressive. It’s a liability.

The question isn’t “can I automate this?” The question is “will I be glad I automated this in 14 months when something breaks at midnight?”

I spent probably six hours over the 18 months debugging automations that nobody in my household had used or noticed. That’s six hours of homelab time I could have spent on literally anything else. The dishwasher notification automation alone cost me four hours to build and zero minutes of value delivered.

There’s also the WAF problem â and before anyone says that acronym is outdated, the principle isn’t. Any automation system that affects other people in your home has to actually work correctly for those people, or it erodes trust in the whole system. My wife now asks me before I add anything new that touches the main floor. That’s a reasonable response to having the lights dim unexpectedly for two months.

The Honest Tradeoffs

Home Assistant is genuinely powerful, and that’s part of the problem

Home Assistant is one of the best pieces of open source software I run. The 2024 releases in particular have been stable in a way that older versions weren’t. But the platform actively encourages you to build more. Every integration you add surfaces more entities. Every entity is an invitation to write a condition. The UI makes it easy to start automations and gives you very little friction to stop and think about whether you should.

I’m not blaming the software. I’m blaming the tendency â which I definitely have â to treat the automation count as a proxy for how sophisticated my setup is. It isn’t. My five working automations are more valuable than the 62 I deleted or disabled.

Local processing is real and matters, but it doesn’t fix design problems

I run everything local. No Nabu Casa, no cloud dependencies for core function. That’s a deliberate choice and I think it’s the right one for a setup like mine. But local processing doesn’t make a poorly-designed automation useful. The dishwasher notification fired locally in under a second and was still completely useless. Speed and reliability are table stakes, not the thing that makes an automation worth having.

Canadian-specific note on presence detection

If you’re in Canada and using phone GPS presence for automations, your carrier’s network geometry matters more than most guides acknowledge. I’m on a smaller regional carrier and my phone’s GPS assist takes noticeably longer to resolve after being indoors, which caused presence detection delays that broke several automations before I added debounce delays. Bell and Rogers customers in dense Calgary neighbourhoods may not see this, but if you’re rural or on an MVNO with a different tower footprint, test your presence detection aggressively before building anything that depends on it.

What I Would Do Differently From Month One

If I were starting over, I would enforce a rule: no automation gets built until it’s been a manual inconvenience at least three times in a month. Not once. Not twice. Three times, where someone in the house said “I wish that happened automatically.” That threshold filters out almost all of the theatre builds immediately.

I’d also keep a plain text log of every automation with three fields: what it does, when it last did it, and whether anyone noticed. Review it quarterly. Anything with “nobody noticed” in the last three months gets deleted unless there’s a strong reason to keep it.

The workshop heat automation, the gate alert, the dehumidifier logic â those would all have survived that filter. The dishwasher notification, the welcome lighting, the good morning routine â none of them would have made it past month two.

There’s no shame in a simple setup. Forty entities and five automations that work reliably beats 140 entities and 67 automations where you can’t explain half of them. I had to learn that the hard way over 18 months and one very humbling audit.

Go open your automations list right now and ask yourself: which of these would I actually miss? Delete the rest. Your future self will thank you when the next Home Assistant update ships and something inevitably breaks.

Related Auburn AI Products

Building a homelab or self-hosting content site? Auburn AI has practical kits:

500 Homelab and Self-Hosting Blog Titles ($27)
Auburn AI Monitoring Stack ($37) – 6 production PowerShell scripts
Podcast Automation Kit ($37)
Browse all Auburn AI products