Over about four years I accumulated a lot of connected devices. Then I spent a weekend removing roughly half of them, and the house works better than it did.
The ones that survived share a property that I did not identify until I was throwing things away.
What I removed
Connected bulbs in rooms where the light switch was perfectly adequate. The failure mode is fundamental: somebody turns the bulb off at the wall and now it is unreachable by any other means, which makes it worse than a bulb.
A voice-controlled kettle, which required more interaction than filling and switching on a kettle.
Sensors that reported information nobody acted on. Humidity in three rooms, generating a graph I looked at twice.
A connected doorbell that produced notifications I stopped reading within a month.
And several devices whose manufacturers had discontinued support, which is its own category and the most annoying one.
What survived
Heating control, which is the clearest win. Scheduling, zoning and turning it off remotely genuinely saves money and genuinely improves comfort, and the manual alternative is worse in every respect.
Smart plugs on things that should not be left on, used for scheduling rather than for voice control.
A leak sensor near the washing machine, which has done nothing for three years and would justify itself entirely on one event.
Outdoor lighting on a schedule, which was previously a timer switch and is now slightly better.
And door sensors, which tell me whether the house is secured without walking around it.
What the survivors have in common
The thing I did not see until afterwards.
Every one of them does something a person cannot do, or does something automatically that would otherwise require remembering.
The ones I removed all replaced a physical action with a different physical action, generally a slower one involving a phone.
That is the whole distinction. Automation that removes a task is valuable. Automation that changes the interface to a task is usually a downgrade, because the physical switch was already very good.
The reliability problem
The practical reason the failures were worse than no automation at all.
A light switch works. It works during an internet outage, when the manufacturer's servers are down, when the application has updated and requires a login, and when a guest who has never seen it before wants light.
A connected bulb fails in all of those cases, and the failure is confusing rather than obvious.
Which means the reliability floor of the house is set by the least reliable component in any given room, and adding a dependency to something that previously had none is a real cost.
The discontinuation risk
The one that produced actual waste.
Several devices depended on a manufacturer's cloud service. When the manufacturer discontinued the product line, the service went with it, and functioning hardware became unusable.
This has happened repeatedly across the industry and it is not an unusual outcome. Hardware companies get acquired, pivot, or fail, and the service disappears.
Which is the strongest argument for devices that work locally, and for open standards, and against anything where the core function requires somebody else's server to be running.
What I would install in a new house
Heating control, first and without hesitation.
Leak and smoke detection, on the basis that the cost is trivial against what they prevent.
Scheduled outdoor lighting.
Smart plugs, sparingly, on specific things.
And nothing that replaces a working physical control with an application, which eliminates most of the category.
The wiring point
A practical note for anybody doing this properly rather than incrementally.
Devices that are wired, and that control the circuit rather than the bulb, avoid the whole class of problem where somebody uses the wall switch.
Smart switches replacing the existing switch keep the physical control working normally while adding remote capability, which is the correct arrangement and the one I should have started with.
It costs more, requires electrical work, and produces a house where nothing is worse than it was before, which is a low bar that most of my original installation failed to clear.
The household test
The check I now apply before installing anything, which came from complaints rather than from analysis.
Can everybody in the house operate it without instruction, including a visitor and somebody who has never used it.
A system that requires knowing which application, which account and which spoken phrase is a system that only works for the person who installed it.
That failed test is why several of the removals happened. The automation was fine and the household had quietly routed around it, which is the clearest possible verdict.
Documenting it
A short written note of what is installed, what it does and how to override it manually is worth writing, particularly for heating.
The situation to avoid is somebody else needing to fix something while you are unreachable, faced with a system nobody has explained.