Argues that a light switch that stops working because of an us-east-1 outage is 'a liability with an app,' not a smart switch. Points to a 15-year pattern of failures — Revolv, Nest API deprecation, Insteon's overnight disappearance, Chamberlain MyQ cutting third-party access, and Belkin WeMo's scheduled 2026 shutdown — as evidence that this is the business model, not isolated incidents.
Contends the engineering cost of a local-control path is negligible, so the persistence of cloud dependency can only be explained by business incentives: recurring revenue from devices that phone home, lock-in, and data collection. Frames the cloud as something that exists for the vendor's benefit, not the customer's.
Argues that Matter 1.3, Home Assistant, ESPHome, and the resurgence of Zigbee and Z-Wave in mainstream retail have made local control practical without requiring hardware hacking. The 'works offline' requirement is now checkable off-the-shelf, which changes the conversation from aspiration to actionable buying criteria.
A post titled "Home automation: an overnight success, 15 years in the making" hit 240 points on Hacker News this week, and the comment section turned into the usual ritual bloodletting over cloud-dependent smart home gear. The author's thesis is simple and, at this point, uncontroversial among people who have actually wired up a house: a light switch that stops working because a server in us-east-1 is having a bad day is not a smart switch — it's a liability with an app.
The specifics in the thread are grimly familiar. Users recounting Nest thermostats that bricked when Google deprecated the Works-with-Nest API. Chamberlain MyQ garage openers that cut off third-party integrations and then started showing ads. Insteon's hub company vanishing overnight in 2022 and taking an entire ecosystem with it. Belkin WeMo's recent announcement that it is ending cloud service for most of its product line in 2026, turning working hardware into e-waste on a scheduled timer. The pattern is not a series of accidents. It is the business model.
What's new is not the complaint — developers have been yelling about this since the original Revolv shutdown in 2016 — but the arrival of a credible local-first alternative stack. Matter 1.3, Home Assistant's continued dominance among tinkerers, ESPHome's maturation, and the quiet resurgence of Zigbee and Z-Wave radios in mainstream retail gear mean the "works offline" box is now checkable without soldering.
The interesting question isn't whether cloud dependency is bad. Everyone in the thread agrees it's bad. The interesting question is why vendors keep shipping it anyway when the engineering cost of a local-control path is negligible. The answer, laid bare in the HN comments by several ex-IoT engineers, is that the cloud isn't there for you. It's there for the business.
Cloud dependency enables three things vendors won't give up easily. First, recurring revenue: a device that phones home can be paywalled, tier-gated, or ad-injected long after you paid for the hardware. Ring, Arlo, and MyQ have all done exactly this. Second, telemetry: every button press, every motion event, every time you open your fridge is a data point someone wants to sell or train a model on. Third, kill-switch leverage: if the vendor decides to EOL the product, brick the previous owner's unit after resale, or force you onto a new subscription, cloud control is the mechanism.
Contrast this with the local-first camp. Home Assistant now has over 500,000 reported active installations according to its 2025 analytics opt-in data. The Zigbee2MQTT project supports north of 4,000 devices. ESPHome lets a $3 ESP32 replace a $40 cloud-locked sensor with roughly four lines of YAML. The technology to build a smart home that survives its vendor's bankruptcy has been shipping, open source, and improving for a decade. The gap is entirely on the demand side.
Matter was supposed to fix this at the protocol layer, and partially has. Matter devices are required to support local control over Thread or Wi-Fi, and the spec explicitly bans mandatory cloud dependency for core functions. But vendors have found the usual workarounds: the device works locally, but the *app* that configures it doesn't; or the advanced features (geofencing, automations, voice) are cloud-gated even though the on/off is local. Reading the fine print on a Matter-logo box in 2026 requires the same paranoia it did in 2020.
The community reaction in the HN thread is worth quoting. One commenter, a former Nest engineer, noted that "every meeting about offline mode ended with someone from product asking how we'd monetize users who never came back online." Another, running a 120-device Home Assistant setup, pointed out that his power bill went *down* after ripping out cloud devices because the constant WAN chatter and always-on radios were measurable. A third summarized the whole genre: "The smart home works. The smart home industry doesn't."
If you're building or buying for a home — your own or a client's — the practitioner playbook has gotten considerably clearer in the last two years. The short version: assume every cloud will eventually go away, and architect accordingly.
For the controller layer, Home Assistant on a Raspberry Pi 5 or an N100 mini-PC is now the default serious answer. It runs locally, has first-class Matter, Zigbee (via a $20 Sonoff or ConBee stick), Z-Wave, and MQTT support, and its automation engine is more powerful than any vendor hub. Alternatives like Hubitat (local-first, commercial, $150 one-time) and OpenHAB exist for specific tastes. If your smart home's brain requires a company's servers to think, you don't own a smart home; you lease one.
For device selection, the questions to ask before buying anything are concrete. Does it expose a local API (REST, MQTT, Matter) with no cloud account required for basic functions? Is it supported by Home Assistant's integration list? Can it be flashed with Tasmota or ESPHome if the vendor disappears? For sensors and switches, Zigbee and Z-Wave gear from Aqara, Sonoff, Inovelli, and Zooz clears all three bars. For cameras, Reolink and Amcrest work entirely over local RTSP. For locks, Schlage Encode Plus with Matter is one of the few mainstream options that actually behaves locally.
For what to avoid: anything that requires an account to complete setup, anything whose advertised features include "AI" with no on-device inference claim, and anything from a company whose previous generation got EOL'd within five years. Ring, Wyze, MyQ, and current-gen Nest all currently fail at least one of these filters.
The trajectory is cautiously optimistic. Matter is slowly tightening its local-control requirements, regulators in the EU are circling the practice of bricking paid-for devices via software update (the "right to functional hardware" conversation is live in Brussels), and the Home Assistant ecosystem crossed a threshold in 2025 where non-technical users can realistically run it with the Green or Yellow appliance. The smart home that works offline is no longer a hobbyist flex; it's a buying standard a careful consumer can meet without writing a single line of YAML. The vendors still haven't gotten the memo, but their customers — and increasingly, their regulators — have.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.